(19)
(11)EP 2 904 514 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
25.03.2020 Bulletin 2020/13

(21)Application number: 13734557.5

(22)Date of filing:  14.06.2013
(51)International Patent Classification (IPC): 
G06F 13/10(2006.01)
G06F 13/40(2006.01)
G06F 9/38(2018.01)
H04L 29/06(2006.01)
G06F 16/9535(2019.01)
G06F 16/2453(2019.01)
G06F 13/362(2006.01)
G06F 9/30(2018.01)
G06F 21/62(2013.01)
G06F 16/22(2019.01)
G06F 16/2455(2019.01)
(86)International application number:
PCT/US2013/045881
(87)International publication number:
WO 2014/055138 (10.04.2014 Gazette  2014/15)

(54)

SEMI-JOIN ACCELERATION

SEMI-JOIN-BESCHLEUNIGUNG

ACCÉLÉRATION DE SEMI-JOINTURE


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(30)Priority: 02.10.2012 US 201261709142 P
26.02.2013 US 201313778013

(43)Date of publication of application:
12.08.2015 Bulletin 2015/33

(73)Proprietor: Oracle International Corporation
Redwood Shores, CA 94065 (US)

(72)Inventors:
  • AINGARAN, Kathirgamar
    Redwood Shores, California 94065 (US)
  • SWART, Garret F.
    Redwood Shores, California 94065 (US)

(74)Representative: D Young & Co LLP 
120 Holborn
London EC1N 2DY
London EC1N 2DY (GB)


(56)References cited: : 
US-A1- 2008 114 724
US-A1- 2008 162 876
  
  • Pranav Vaidya: "A Novel Multicontext Coarse-Grained Join Accelerator For Column-Oriented Databases", ERSA09 , 2009, XP002711841, Retrieved from the Internet: URL:http://www.learningace.com/doc/3226820 /afb1fbaf074fbb0afaf802a19c86571c/ersa09_j oin_accelerator [retrieved on 2013-08-27]
  • PRANAV VAIDYA ET AL: "A Novel Multicontext Coarse-Grained Reconfigurable Architecture (CGRA) For Accelerating Column-Oriented Databases", ACM TRANSACTIONS ON RECONFIGURABLE TECHNOLOGY AND SYSTEMS, vol. 4, no. 2, 1 May 2011 (2011-05-01), pages 1-30, XP55076632, ISSN: 1936-7406, DOI: 10.1145/1968502.1968504
  
Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


Description

PRIORITY CLAIM AND RELATED APPLICATION



[0001] This application claims priority to U.S. Patent Application No. 13/778,013, filed February 26, 2013; which claims priority to U.S. Provisional Application No. 61/709,142, filed October 2, 2012.

[0002] This application is related to U.S. Patent Application No. 13/778,009 filed February 26, 2013.

FIELD OF THE INVENTION



[0003] The present invention relates generally to processing a query and, more specifically, to using custom hardware in one or more coprocessors to perform one or more operations that are required to process the query.

BACKGROUND



[0004] The approaches described in this section are approaches that could be pursued, but not necessarily approaches that have been previously conceived or pursued. Therefore, unless otherwise indicated, it should not be assumed that any of the approaches described in this section qualify as prior art merely by virtue of their inclusion in this section.

[0005] Queries issued to a database typically target one or more one or more database objects, such as relational tables. Many times, accessing data organized in a relational table involves scanning the relational table or at least a portion thereof. A common SQL query is one that requires a filter on a database table, such as the following:
select EMPLOYEE from T_EMPLOYEES where HIRE_YEAR= '2012'

[0006] In this example, that database table T_EMPLOYEES is searched for all the employees who were hired in 2012. This search (or "scan") is done by software running on one or more microprocessors that execute a series of instructions to search through the table for the specified value, which is '2012' in this example. The first step is typically the performance bottleneck when running analysis applications on a large database, since this step has to run on the entire table, which may be several terabytes large. Subsequent steps will work on the filtered subset of the first scan step that meets the criteria set in the scan (employees hired in 2012 in the above example). Therefore, the number of rows that a machine can filter per unit of time is an important performance metric for the machine. This metric is referred to as the "scan rate."

[0007] Approaches for processing queries, such as queries that involve scanning a table, have relied on software techniques, where the software is executed (or "runs") on a general purpose microprocessor.

[0008] Pranav Vaidya: "A Novel Multicontext Coarse-Grained Join Accelerator For Column-Oriented Databases", 2009, XP002711841 discloses that reconfigurable logic coprocessors may be a promising solution for hardware acceleration of column-oriented
databases. The paper describes a multicontext, coarse-grained reconfigurable coprocessor unit that can be used to accelerate the join operator in column-oriented databases.

[0009] Pranav Vaidya et al: "A Novel Multicontext Coarse-Grained Reconfigurable Architecture (CGRA) For Accelerating Column-Oriented Databases", ACM TRANSACTIONS ON RECONFIGURABLE TECHNOLOGY AND SYSTEMS, vol. 4, no. 2, 1 May 2011, pages 1-30, XP055076632 discloses that the storage model of column-oriented databases is similar in structure to densely packed matrices/vectors found in many high-performance computing applications. Hence, hardware-accelerated vectorized matrix operations using reconfigurable logic coprocessors may find parallels in hardware acceleration of databases. A multicontext, coarse-grained reconfigurable coprocessor unit model is proposed that is used to accelerate some of the database operations in hardware for column-oriented databases. The implementation of hardware algorithms id described for the equi-join, nonequi-join, and inverse-lookup database operations.

[0010] US 2008/114724 A1 discloses a method and system for integrating an enterprise's structured and unstructured data to provide users and enterprise applications with efficient and intelligent access to that data. Queries can be directed toward both an
enterprise's structured and unstructured data using standardized database query formats such as SQL commands. A coprocessor can be used to hardware-accelerate data processing tasks (such as full-text searching) on unstructured data as necessary to handle a query. Furthermore, traditional relational database techniques can be used to access structured data stored by a relational database to determine which portions of the enterprise's unstructured data should be delivered to the coprocessor for hardware-accelerated data processing.

SUMMARY



[0011] The invention is defined by the independent claims. Further aspects of the invention are outlined in the dependent claims.

BRIEF DESCRIPTION OF THE DRAWINGS



[0012] In the drawings:

FIG. 1 is a block diagram that depicts an example computer system for accelerating a table scan, according to an embodiment;

FIG. 2 is a block diagram that depicts an example coprocessor, according to an embodiment;

FIG. 3 is a flow diagram that depicts a process for processing a query, according to an embodiment;

FIG. 4 is a block diagram that depicts a portion of an example lookup vector, in an embodiment; and

FIG. 5 is a block diagram that illustrates a computer system upon which an embodiment of the invention may be implemented.


DETAILED DESCRIPTION



[0013] In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, that the present invention may be practiced without these specific details. In other instances, well-known structures and devices are shown in block diagram form in order to avoid unnecessarily obscuring the present invention.

GENERAL OVERVIEW



[0014] In an embodiment, a scan operation or a lookup operation associated with a query is implemented in hardware, such as a coprocessor that is located on the same chip as a general purpose microprocessor. In this way, the scan operation is performed by custom hardware whereas other portions of the query are handled by the general purpose microprocessor running software. One advantage of having different hardware components perform different operations is that custom hardware is better able to handle the large volume of data that is required for a scan operation or a lookup operation. Also, custom hardware frees up the general purpose microprocessor and associated cache(s) to work on other parts of the query or even other tasks that are unrelated to the software that processes the query.

[0015] Embodiments of the invention are not limited to any particular microprocessor or graphic processing unit (GPU).

[0016] The following examples refer to a table as a data object that is scanned. However, all embodiments are not limited to tables. Data objects other than tables may be scanned.

SYSTEM OVERVIEW



[0017] FIG. 1 is a block diagram that depicts an example computer system 100 for accelerating the processing of a query, according to an embodiment. Computer system 100 includes a query execution engine 110, an OS/hypervisor 120, a coprocessor 130, and memory 140. Although only a single coprocessor 130 is depicted, system 100 may include multiple coprocessors.

[0018] Query execution engine 110 and hypervisor 120 are programs that reside in memory (e.g., DRAM and/or cache memory) and include instructions that are executed by a general purpose microprocessor. Query execution engine 110 comprises one or more software components and may communicate with one or more other software components that are not part of query execution engine 110 in order to execute a query to generate a result of the query. Query execution engine 110 may be configured to rewrite a query (e.g., a SQL query) to generate a rewritten query that query execution engine 110 is able to execute. Alternatively, another software component receives an original query, generates a rewritten query based on the original query, and passes the written query to query execution engine 110 for processing.

[0019] Non-limiting examples of types of queries that query execution engine 110 may be configured to process include SQL queries and XML queries, such as XPath queries and XQuery queries. At least one type of query that query execution engine 110 is configured to process is a query that requires a scan of an object, or a portion thereof. As noted previously, a non-limiting example of an object that is scanned is a relational table that is logically organized in one or more columns and multiple rows. While data may be logically organized in a single table, the data may be organized very differently in persistent storage, such as a hard disk drive or a flash memory device. For example, the data of a table may be partitioned or different columns of a table may be stored in very different storage locations.

[0020] Hypervisor 120 acts as an interface between query execution engine 110 and coprocessor 130. In other words, commands issued by query execution engine 110 to coprocessor 130 are issued over hypervisor 120. Thus, query execution engine 110 issues commands on the hypervisor interface by making API calls to OS/hypervisor 120.

[0021] A hypervisor is a hardware virtualization technique that allows multiple operating systems ("guests") to run concurrently on a host computer. A hypervisor presents, to guest operating systems, a virtual operating platform and manages the execution of the guest operating systems. Multiple instances of a variety of operating systems may share the virtualized hardware resources. Hypervisors may be installed on server hardware, with the function of running guest operating systems that, themselves, act as servers.

[0022] One type of hypervisor runs directly on a host's hardware to control the hardware and to manage guest operating systems. A guest operating system thus runs on another level above the hypervisor. Another type of hypervisor runs within a typical operating system environment. With the hypervisor layer as a distinct second software level, guest operating systems run at the third level above the hardware. In other words, the first type of hypervisor runs directly on the hardware while the second type of hypervisor runs on another operating system, such as FreeBSD, Linux, or Windows.

[0023] Thus, although element 120 is labeled as "OS/Hypervisor," an operating system and a hypervisor are different entities. For purposes of this description, an OS and hypervisor are treated the same. The following references to element 120 will be "hypervisor 120."

[0024] Although hypervisor 120 is depicted as part of computer system 100, in one embodiment, computer system 100 does not include a hypervisor. In that embodiment, query execution engine 110 issues commands directly to coprocessor 130 without first requiring processing by any other software component, other than an operating system (not depicted) of computer system 100.

[0025] Coprocessor 130 is a hardware element that is programmed to perform one or more tasks separate from the tasks performed by the general purpose processor that executes query execution engine 110 and hypervisor 120. While coprocessor 130 is separate from the general purpose processor that executes query execution engine 110, coprocessor 130 may be viewed as part of query execution engine 110 in that coprocessor 130 performs one or more tasks that were previously performed by query execution engine 110.

[0026] In an embodiment, coprocessor 130 at least performs the task of comparing a specified target value (or target range of values) against a series of input values from a table. This task is referred to as a scan operation and is described in more detail below. In an embodiment, coprocessor 130 is capable of comparing multiple specified target values (or multiple specified target ranges) against a series of input values or data elements from a table.

[0027] In another embodiment, coprocessor 130 at least performs the task of determining whether one or more values exist in a particular set of values. The one or more values may be used to index into the particular set of values so that the particular set of values do not need to be scanned for each value of the one or more values. In this embodiment, coprocessor 130 is programmed to efficiently perform a lookup operation, which is described in more details below.

[0028] Other than initial parameters established or dictated by query execution engine 110, coprocessor 130 performs a scan operation and/or a lookup operation without intervention from query execution engine 110 or any other software until the scan operation or lookup operation (indicated by the initiating command) completes. At that point, coprocessor 130 signals to query execution engine 110 that results of the operation are available. The signal may be in the form of setting a flag. Hypervisor 120 may use this signal to insert new commands into a command queue of coprocessor 130.

[0029] In an embodiment, coprocessor 130 is programmed to handle different data types/formats and element sizes. For example, coprocessor 130 may process data that is in a string format, a date format, or a number (e.g., integer or float) format. Also, the size of a data element that coprocessor 130 processes may be a particular number of bits (e.g., 7 bits) or a particular number of bytes (e.g., 2-bytes). Furthermore, data elements from a particular source (such as a table) may be variable length or fixed length. In an embodiment, a data element that coprocessor 130 receives from an object (such as a table) may be of one size and coprocessor 130 performs an operation to reduce or increase the size of the data element, such as removing one byte from the data element, adding 9 bits to the data element or decompressing the data element, before performing, for example, a comparison of the data element with another data element.

[0030] In an embodiment, coprocessor 130 resides on-chip, that is, on the same chip as a general purpose microprocessor that executes query execution engine 110. Coprocessor 130 includes (a) a memory interface that streams table data (or other data) from on- or off-chip memory to coprocessor 130 and (b) a compute block that performs a scan operation and/or a lookup operation. For example, in the case of the scan operation, the compute block acts on the table data to determine if a specified value, or a range of values, occurs in the table data. Thus, a set of comparators is used to determine if each element of an incoming stream is equal to a searched-for value or lies in a searched-for range of values. Each comparator in the set of comparators may perform a comparison operation at the same time. Thus, coprocessor 130 may perform multiple comparison operations simultaneously. In an embodiment, coprocessor 130 is configured to perform multiple types of comparisons, such as one 4-byte comparison, two 2-byte comparisons, four 1-byte compares, and/or one 2-byte and one 1-byte comparison. Coprocessor 130 sends results of a search to on-chip memory or off-chip memory through the memory interface.

[0031] As depicted in FIG. 1, memory 140 stores commands 142, input data 144, and output data 146. A command reflected in commands 142 refers to (1) a location in memory 140 that stores at least a portion of input data 144 and (2) a location in memory 140 that results (generated by coprocessor 130) of the operation(s) that correspond to the command will be stored.

[0032] In an embodiment, coprocessor 130 includes a command queue that stores one or more addresses of one or more commands. When not busy, coprocessor 130 selects one or more addresses (inserted by hypervisor 120) from the command queue in order to retrieve the one or more commands (e.g., reflected in commands 142) from memory (e.g., memory 140).

COPROCESSOR CONTROL BLOCK



[0033] In an embodiment, query execution engine 110 includes instructions that, when executed by a general purpose microprocessor (not depicted), causes generation of a coprocessor control block (CCB). A CCB is a data structure that represents a command issued by query execution engine 110 and that includes data that coprocessor is configured to read and process. In an embodiment, a CCB includes command type data that indicates the type of operation coprocessor 130 is to perform and one or more operands that correspond to the operation indicated by the command type data. If coprocessor 130 only performs one operation, then command type data may not be an available operand in a CCB. Alternatively, coprocessor 130 may ignore the command type data if coprocessor is configured to perform only one operation.

[0034] The command type data indicates which logic coprocessor 130 will use to process a command. Thus, different command types correspond to different logic implemented by coprocessor 130. For example, a scan operation requires coprocessor 130 to execute first logic while a lookup operation requires coprocessor 130 to execute second logic that is different than the first logic.

[0035] An operand indicated in a CCB may be one of two types: an immediate operand or an indirect operand. An immediate operand is an operand that can be used immediately by a coprocessor when the coprocessor performs the operation without first requiring translation of the operand, such as a memory lookup. An example of an immediate operand in the context of a scan operation is a 4-byte integer that is used to perform a comparison against data elements from table data. An indirect operand is an operand that must first be translated or looked up before the coprocessor can perform the designated operation. An example of an indirect operand is a physical address that indicates where (e.g., in memory 140) table data is stored for the coprocessor to perform the operation, whether a scan operation or a lookup operation.

[0036] In the context of a scan operation, the operands (of the scan operation) indicated in a CCB include (a) comparison data that indicates data that is used to perform a comparison against data from table data and (b) location data that indicates where the table data is located (e.g., input data 144 in memory 140).

[0037] Comparison data may be any type of data, such as a number, a date, a character, or a string. Comparison data may be a single value and/or a range of values. Additionally, comparison data may indicate multiple values and/or multiple ranges of values. For example, a query may request to view names of employees who make below $30,000 and employees who make between $100,000 and $130,000. In this example, comparison data indicates a range of 0-30,000 and a range of 100,000-130,000.

[0038] Location data may be a single address or a multiple addresses, such as a starting address and an ending address or a starting address and an offset from the starting address. Each address indicated in the location data may be a virtual address, a real address, or a physical address. In an embodiment, hypervisor 120 replaces location data indicated in a CCB with second location data. For example, a (e.g., guest) operating system identifies the location data indicated in a CCB, where the location data is a virtual address and replaces the virtual address with a real address. The operating system then sends the CCB to hypervisor 120. Hypervisor 120 looks up, in a mapping table, a physical address that is mapped to the real address, and replaces, in the CCB, the real address with the physical address.

[0039] In an embodiment, a CCB also includes output location data that indicates where coprocessor 130 is to send a result of performing an operation indicated by the CCB. In FIG. 1, the output location data would point to output data 146 in memory 140. This may be helpful if multiple microprocessors are integrated on the same chip and each microprocessor has its own private cache. Thus, if a particular general purpose microprocessor that executes query execution engine 110 is one of multiple general purpose microprocessors on the same chip and each is microprocessor is associated with different (e.g., L3) cache, then query execution engine 110 may specify, as a parameter in a CCB, a cache that is adjacent to or near the particular general purpose microprocessor. Thus, instead of coprocessor 130 sending results of an operation to RAM, coprocessor 130 may send (based on output location data indicated in the CCB) the results not only to cache, but to a specific cache that is "closest" to query execution engine 110. In this way, query execution engine 110 is not required to request the results (a) from RAM, (b) from another microprocessor's (or core's) cache, or (c) from shared cache that is shared among multiple cores, each which may be much slower than accessing data from a microprocessor's own (private) cache. Instead, query execution engine 110 is allowed to dictate where results of operations performed by one or more hardware elements (i.e., coprocessors in this embodiment) will be stored.

COPROCESSOR



[0040] Once coprocessor 130 receives a command (e.g., in the form of a CCB) over an interface of hypervisor 120 (or directly from query execution engine 110), coprocessor 130 executes the command asynchronous to the thread of query execution engine 110 issuing the original command. If coprocessor 130 receives multiple commands, then coprocessor 130 may schedule the multiple commands for execution in a round robin fashion. Some commands may be executed in parallel.

[0041] In an embodiment, input data (e.g., relational data) for a command is fetched over an interface (to query execution engine 110) and results of a command (i.e., results that coprocessor 130 generates based on the input data) are written out over the interface.

[0042] In an embodiment, coprocessor 130 causes a completion status to be written out, over the interface at the end of each command, to a completion data structure in the interface. Query execution engine 110 may use the completion data structure to resynchronize with one or more threads of query execution engine 110.

[0043] FIG. 2 is a block diagram that depicts an example coprocessor 200, according to an embodiment. Coprocessor 200 may be coprocessor 130 in FIG. 1. Coprocessor 200 includes a memory interface 210, a command scheduler 220, a decompressor 230, query pipe 240, and a message pipe 250. Message pipe 250 handles memory copies and message passing. Query pipe 240 handles one or more query commands, such as a scan command or a lookup command, after decompressor 230 decompresses compressed input data (e.g., data from a relational table).

[0044] Decompressor 230 may be configured to decompress only data that is compressed in a single format. Alternatively, decompressor 230 may be configured to decompress data that is compressed in one format and other data that is compressed in another format. In an embodiment, coprocessor 200 does not include decompressor 230. Decompression may not be necessary if the data that coprocessor 200 receives is not compressed (e.g., is already decompressed) when coprocessor 200 receives the data. Also, decompression may not be necessary even for compressed data if coprocessor 200 is configured to operate directly on the compressed data without having to first decompress the compressed data.

[0045] Each of pipes 240 and 250 is associated with a different set of command queues and, optionally, command formats. Hypervisor 120 is configured to ensure that commands (reflected in CCBs) are directed to the command queue of the correct pipe. A flag bit in a CCB may indicate if the CCB is a message command or a query command.

[0046] Each of pipes 240 and 250 may be multithreaded and capable of executing multiple commands at a time. The degree of multithreading is not exposed to software. Command scheduler 220 may schedule the commands on available threads on the assumption the commands are parallelizable. If a given command needs to be serialized behind another command, then the two commands may be placed in the same command queue and the appropriate serializing flags may be set in both commands.

[0047] Although not depicted, coprocessor 200 comprises a certain amount of memory to store data as the data is streamed through memory interface 210 or to store data that is used in a lookup operation, such as a lookup vector, an example of which is a Bloom filter. The size of the memory of coprocessor 200 may be quite small (e.g., 4KB) due to modern chips that consist largely of one or more caches for the main core(s) or general purpose microprocessor(s).

[0048] While query execution engine 110 "views" table data relational and performs operations as such, coprocessor 200 only "sees" or operates on vectors or single dimensional arrays of data. In other words, coprocessor 200 does not "view" multiple columns or row identifiers. Rather, coprocessor 200 is agnostic when it comes to how the data is logically organized or stored. Therefore, in providing instructions to coprocessor 200, query execution engine 110 ensures that the output of any operations performed by coprocessor 200 is stored in a particular order. If not, the query execution engine 110 would not know which portion of the table to which the output corresponds. One way in which ordering is preserved is for query execution engine 110 to keep track of which set of table data corresponds to which CCB, where each CCB includes a unique CCB identifier. Then, the output generated by coprocessor 200 based on a particular CCB includes the identifier for that particular CCB to allow query execution engine 110 to determine to which portion of the logical table the output corresponds. For example, query execution engine 110 may store association data that associates rows 1001-2000 of table Employee with CBB identifier 432899. Coprocessor 200 receives and processes a CCB with identifier 432899 to generate output that is stored at a certain location.

[0049] Alternatively, instead of keeping track of a CCB identifier, query execution engine 110 stores association data that associates table data that indicates a portion of a table (e.g., rows 1001-2000 of table Employee) with output location data (e.g., physical address 1298737+4KB) that indicates where output generated by coprocessor 200 is to be stored. Later, when query execution engine 110 examines the output stored at that storage location, query execution engine 110 uses the association data to determine which portion of the table corresponds to that output. Thus, query execution engine 110 can keep track of the order of the output even though coprocessor 200 operates on different portions of the table at different times and even though query execution engine 110 might instruct multiple coprocessors to operate on different portions of the table, which operations might be performed concurrently.

PROCESSING A QUERY



[0050] FIG. 3 is a flow diagram that depicts a process 300 for processing a query that requires a scan operation, in an embodiment. At block 310, query execution engine 110 receives a query that targets one or more data objects, such as a table, and that requires a scan operation of at least one of the one or more data objects. For example, query execution engine 110 may process a SQL query to generate a rewritten query that includes one or more database operations, including a scan operation, that query execution engine 110 is configured to execute or to instruct one or more other software components to execute. Alternatively, another software component receives an original query and generates a rewritten query that query execution engine 110 is configured to process.

[0051] At block 320, query execution engine 110 determines, based on one or more criteria, whether to involve coprocessor 130 in processing the query. The one or more criteria may indicate whether the result was previously generated and cached, whether an index on the table exists and may be used to answer the query instead of scanning the table, the size of the table, etc. For example, if the size of the table that needs to be scanned is relatively small, then involving coprocessor 130 may require more work (e.g., in the form of usage of the general purpose microprocessor that is executing the instructions of query execution engine 110) or take more time than executing the query without involving coprocessor 130.

[0052] Additionally or alternatively, the one or more criteria may indicate a relative cost for processing the query (or rewritten query) in different ways. For example, query execution engine 110 may include a cost estimator component that estimates the cost of executing the query under different execution plans, such as using an index, scanning the table without using coprocessor 130, and scanning the table using coprocessor 130. Query execution engine 110 then selects the execution plan that is the least expensive in terms of cost. "Cost" may be based on one or more factors, such as CPU usage, memory usage, I/O usage, and network I/O usage.

[0053] If query execution engine 110 determines to involve coprocessor 130 in executing the query, then process 300 proceeds to block 330.

[0054] At block 330, query execution engine 110 sends, to hypervisor 120, an address of the one or more instructions, an address of the input data, and an address of where output data is to be stored. The one or more instructions may be in the form of a CCB that query execution engine 110 generates. Hypervisor 120 translates the addresses from virtual addresses into physical addresses and places the physical addresses into a command queue or buffer of coprocessor 130.

[0055] After query execution engine 110 causes the one or more instructions to be stored in memory (and, thus, are available for coprocessor 130 to read), query execution engine 110 may perform other tasks that are related to execution of the query or that are related to another query altogether. In this way, the operation(s) performed by coprocessor 130 are performed asynchronously to the tasks performed by query execution engine 110, which is executed by a general purpose microprocessor.

[0056] As noted above, computer system 100 may include multiple coprocessors. Thus, query execution engine 110 may send instructions (e.g., a CCB) to each of multiple coprocessors. In this way, a scan operation or a lookup operation may be divided up into multiple "mini" operations, allowed each coprocessor to perform a different "mini" operation. For example, a particular table may comprise 10,000 rows and there may be ten coprocessors. Query execution engine 110 may then generate ten different CCBs, each of which is similar to the other CCBs except that each CCB indicates a different address from which to access a different set of 1,000 rows from the particular table. In this way, the ten coprocessors operate in parallel on a different portion of the particular table.

[0057] Additionally or alternatively, block 330 involves query execution engine 110 selecting, based on one or more criteria, a subset of multiple coprocessors to send a CCB. For example, query execution engine 110 may only need three coprocessors of ten total coprocessors to each perform a scan operation (but on a different set of table data relative to each other coprocessor). The one or more criteria that query execution engine 110 uses to select one or more coprocessors may be a current load of each coprocessor, latency of each coprocessor, and/or processing history of each coprocessor. For example, query execution engine 110 selects the three coprocessors that are currently the least "loaded" or busy. The load of a coprocessor may be reflected in the number of commands that are in one or more command queues of the coprocessor. Thus, the more commands that are waiting to be processed by a particular coprocessor, the more loaded that particular coprocessor becomes.

[0058] At block 340, coprocessor 130 receives the one or more instructions and performs one or more operations reflected in the one or more instructions. For example, coprocessor 130 receives a CCB, determines that the type of operation(s) reflected in the CCB, reads in any data necessary to complete the operation(s), performs the operation(s), and (in block 350) causes results of the operation(s) to be sent to query execution engine 110. Execution of a command by coprocessor 130 may be triggered by a write, by query execution engine 110 (or one of its agents), to one or more internal registers of coprocessor 130.

SCAN OPERATION



[0059] In an embodiment, the one or more instructions indicate a scan operation and one or more addresses where table data is stored. Coprocessor 130 retrieves the table data and performs comparisons between a value or range of values (specified in the one or more instructions) and the table data. Coprocessor 130 requests the table data from query execution engine 110 through memory (e.g., memory 140), which may be dynamic RAM in the system or cache memory on the chip. Table data may be stored in blocks, which may be relatively large, such as 64KB or larger. Coprocessor 130 may access each of these blocks as a single dimensional array. In a columnar database, data is in a single dimensional array and easily readable by coprocessor 130. In a row major database, data may be first transposed into a column major format before the data is processed by coprocessor 130.

[0060] If the table data spans blocks that are discontinuous in memory, then coprocessor 130 separately requests each block (as a separate job). In such a scenario, query execution engine "stitches" together the results (generated by coprocessor 130) of each job. For example, in a row major database, the data will be strided and coprocessor 130 will select every Nth piece of data where N is specified in the command.

[0061] Coprocessor 130 may perform the comparisons "on-the-fly"; that is, as the table data is streamed to coprocessor 130. Once a data element in the table data is compared to a target value or a target range of values specified in the one or more instructions, coprocessor 130 may (immediately or eventually) overwrite the memory used to store that data element with a new data element from the table data.

[0062] Examples of types of comparison operations that coprocessor may be configured to perform include greater-than (>), less-than (<), equal (==), not equal (!=), greater-than-or-equal-to (>=), and less-than-or-equal-to (<=).

SEMANTIC-AWARE COMPRESSION



[0063] In an embodiment, coprocessor 130 is configured to operate on compressed data. Some data is compressed using one or more non-semantic-aware compression techniques, while other data may be compressed using one or more semantic-aware compression techniques. Data that is compressed using a non-semantic-aware compression technique requires decompression first before the decompressed data may be operated on. Data that is compressed using a semantic-aware compression technique may not need to be decompressed before an operation (for example, a number or string comparison) is performed. An example of a semantic-aware compression technique is run-length encoding (RLE).

[0064] RLE is a form of data compression in which runs of data (that is, sequences in which the same data value occurs in many consecutive data elements) are stored as a single data value and count, rather than as the original run. This is most useful on data that contains many such runs. For example, a column of a table may contain the following sequence department identifiers:
AAAAAABBBCCCCCCCCCDDDDDAAAA

[0065] Applying a RLE data compression algorithm to the above sequence might yield the following output: 6A3B8C5D4A. This run length code represents the original 26 characters in only 10 characters. In RLE, the longer the run of a single data value in an input sequence, the greater the compression.

[0066] Returning to block 340, table data may be run length encoded. Thus, the number of table data that needs to be read into coprocessor 130 and the number of comparisons that coprocessor 130 needs to perform against the run length encoded table data may be substantially less than if the table data is not run length encoded. Given the example above, instead of performing 26 comparisons (i.e., one for each of the 26 characters), coprocessor 130 would only have to perform 5 comparisons.

[0067] The result of performing a scan operation against run length encoded data may itself be run length encoded, which may be eventually processed by query execution engine 110. Given the example above, the result of determining whether a row of a particular table includes department identifier 'A' may be 6Y16N4Y, where 'Y' indicates a positive result of the determination and 'N' indicates a negative result of the determination.

[0068] Alternatively, the result of performing a scan operation against run length encoded data may not be run length encoded. Instead, the result may be "decompressed." Given the example above, the result of determining whether a row of a particular table includes department identifier 'A' may be YYYYYYNNNNNNNNNNNNNNNNYYYY. In this embodiment, although a single comparison is performed for the character 'C' during the scan operation, coprocessor 130 generates eight negative indications (e.g., 'N' or '0') for that run length encoded data element.

LOOKUP OPERATION



[0069] SQL queries frequently need to cross reference multiple tables in a database. Processing such queries typically involves a set-intersect operation. Currently, set-intersect operations are performed by software running on general purpose microprocessors where the software utilizes a vector lookup (e.g., a Bloom filter lookup) when the cardinality of the table columns being joined is small. According to an embodiment, a vector lookup (or lookup operation) is implemented in hardware, which may be much fast than a software implementation.

[0070] Thus, in an embodiment, the one or more instructions (of block 330) indicate a lookup operation, one or more addresses where table data is stored, and one or more addresses where a lookup vector or array is stored. The one or more addresses where the table data (or lookup vector) is stored may be two addresses (e.g., a starting address and an ending address) or a single address with an offset. Coprocessor 130 causes the lookup vector and the table data to be sent to coprocessor 130 and, for each data element in the table data, performs a lookup of the data element in the lookup vector. In other words, coprocessor 130 uses the data element (or a hash of the data element) to identify a position in the lookup vector and retrieve data from the lookup vector at that position.

EXAMPLE QUERY REQUIRING A LOOKUP OPERATION



[0071] An example of a query that may require a lookup operation is a query that requests information about "poor" people who live in "rich" zip codes. A "poor" person may be considered someone who makes less than $30,000 per year and a "rich" zip code may be considered a zip code where the median salary is over $100,000. In this example, coprocessor 130 requires data from at least two data objects: a lookup vector and a Person table. The lookup vector indicates (for example, with a single bit) whether a zip code is "rich" or not. The lookup vector may be pre-computed (i.e., that is, before the query is received) or may be computed in response to receiving the query.

[0072] The Person table contains information about numerous people where the table comprises at least three columns: one for each person's name, one for each person's salary, and one for each person's zip code. In one aspect, the data that indicates whether a zip code is rich or poor is a dimension table while the Person table that contains information about each person is a fact table. Dimension tables are typically much smaller than fact tables. The lookup vector is generated based on the dimension table. Relatedly, the fact table may be represented by multiple data objects (e.g., tables): one data object may contain information about each person's salary and another data object may contain information about each person's zip code.

LOOKUP VECTOR



[0073] An example of a lookup vector is a Bloom filter, which is a probabilistic data structure that is used to test whether an element is a member of a set. While false positives are possible when utilizing a Bloom filter, false negatives are not. A Bloom filter is associated with one or more hash functions, each of which maps an element to one of the array positions in the Bloom filter.

[0074] However, a lookup vector need not be probabilistic. For example, if there are only 10,000 possible zip codes and each zip code is associated with a single bit, indicating whether that zip code is "rich," then the size of the lookup vector (e.g., 1.25KB) may be small enough to fit the entire lookup vector in memory (e.g., SRAM) of coprocessor 130. Therefore, a probabilistic lookup vector is not necessary in order to reduce its size.

[0075] FIG. 4 is a block diagram that depicts a portion of an example lookup vector 400. Each position in lookup vector 400 is associated with a different zip code. In other words, a zip code is used to index into lookup vector 400. Each position in lookup vector 400 contains a single bit indicating if the corresponding zip is a "rich" zip code ('1') or a "poor" zip code ('0').

[0076] If a lookup vector is not capable of fitting entirely in memory of coprocessor 130, then (other than generating a probabilistic lookup vector) either (a) the lookup vector may be split up (or divided) such that coprocessor 130 reads in table data each time for each portion of the lookup vector or (b) a different coprocessor might read in table data once but only for a portion of the lookup vector that it stores, in an embodiment where there are multiple coprocessors.

[0077] In the former scenario, if coprocessor 130 can only fit, for example, ¼ of a lookup vector into its memory, then coprocessor 130 reads table data that indicates a person's zip code four times (i.e., once for each portion of the lookup vector that coprocessor 130 reads in). If a person's zip code is not identified in any one pass of the table data, then the result for that person may indicate a negative determination. The result of each pass of the table data may be one long array of bits, one for each person indicated in the Person table.

[0078] In the latter scenario, four different coprocessors may store a different quarter of the lookup vector and read in once table data that indicates a person's zip code and then perform a lookup into the lookup vector for each zip code reflected in the table data. Again, the result of a lookup operation from each coprocessor may be one long array of bits, one for each person indicated in the Person table.

GENERATING RESULTS OF SCAN OR LOOKUP OPERATION



[0079] In an embodiment, coprocessor 130 generates a specific output format as a result of performing the one or more operations reflected in the one or more instructions. An example of the specific output format is a bit vector, where each position in the bit vector indicates either a true or a false. For example, if "10" is a target value and the comparison is determining whether the target value is greater than a data value or data element from a table, then a result of the comparison would be (a) true if the data value is greater than 10 and (b) false if the data value is less than or equal to 10.

[0080] Each position in the vector corresponds to a data value or data element that was received from the input (e.g., table) data. For example, in the above lookup operation example, coprocessor 130 generates a bit vector that reflects "poor" people that live in "rich" zip codes. In order to generate the bit vector, coprocessor 130 reads in data from the zip code column of the Person table and determines, for each person indicated in the read-in data and based on the lookup vector, whether the person lives in a "rich" zip code. Each bit in the bit vector indicates whether a different person in the Person table lives in a "rich" zip code. Coprocessor 130 later passes the bit vector to the general purpose microprocessor, which uses the bit vector to identify persons that are also considered "poor." For example, for each person that lives in a rich zip code (as indicated in the bit vector), the general purpose microprocessor looks up a corresponding row in the Person table to determine if the person is "poor."

[0081] Because a bit vector is relatively small in size, processing of the bit vector by a general purpose microprocessor is relatively fast; much faster than the general purpose processor processing the input data directly. Furthermore, cache memory space required to store the bit vector is much less than cache memory space that would be required to store the input data (such as a large column of data).

STORING RESULTS OF SCAN OR LOOKUP OPERATION



[0082] Returning to process 300, at block 350, coprocessor 130 causes results of the scan operation (or the lookup operation) to be available to query execution engine 110. Block 350 may involve coprocessor 130 sending the result of an operation to memory that is specified in the one or more instructions from query execution engine 110 that initiated the operation. For example, query execution engine 110 generated a CCB and indicated, in the CCB, that the result of the corresponding operation is to be sent to, for example, DRAM, shared L3 cache, or cache of a specific microprocessor (e.g., that executes query execution engine or that is different than the microprocessor that generated the original command(s)).

[0083] Block 350 may further involve coprocessor 130 setting a flag that, when set, indicates that the operation is complete. This flag setting acts as a signal to (1) hypervisor 120 to insert new requests into a command queue of coprocessor 130 and (2) query execution engine 110 to retrieve the results.

[0084] Alternatively, coprocessor 130 is programmed to always send the result of an operation to a specific memory component (whether RAM or cache) that is accessible to query execution engine 110.

[0085] Alternatively, instead of making the results of the scan operation available to query execution engine 110, coprocessor 130 sends the results to another coprocessor. As noted previously, computer system 100 may comprise multiple coprocessors. The multiple coprocessors may be connected in a series. Each connection between two coprocessors may include a FIFO buffer so that a "producing" coprocessor may generate results faster than a "consuming" coprocessor can process the results. Eventually, the last coprocessor in the "chain" sends results to a specified destination, such as in RAM or cache, that is accessible to query execution engine 110.

[0086] At block 360, query execution engine 110 processes the results and performs one or more other operations in order to generate a final result of the original query. While coprocessor 130 performs the scan or lookup operation, the general purpose microprocessor that executes query execution engine 110 may be idle or may be utilized by query execution engine 110 or another process altogether. For example, query execution engine 110 may perform one or more other operations that are required by the query or that are not be related in any way to the query, but rather to another query.

[0087] As an example of a scan operation, a query might request the IDs and prices of purchase orders that were initiated during a specific range of dates. In this example, coprocessor 130 performs a scan operation that involves reading in date information for multiple purchase orders, where the date information is reflected in a Purchase Order table. The result of the scan operation may be a series of bits (e.g., a bit vector) that each reflects whether a corresponding purchase order was initiated during the specified date range. Coprocessor 130 sends the result to memory that is accessible to query execution engine 110 and may notify query execution engine 110 of the completion of the scan operation by setting a flag that query execution engine 110 checks periodically. Query execution engine 110 then uses the bits to identify, in the Purchase Order table, the entries that correspond to those purchase orders that were initiated during the specified date range. Then, query execution engine 110 identifies the IDs and the prices in the identified entries and returns (e.g., displays) that information as a result of the query. The query may also specify that the result of the query is to be ordered by price in descending order. Thus, query execution engine 110 performs one or more operations after receiving the result of the scan operation performed by coprocessor 110.

[0088] Given the lookup operation example where the query is to identify "poor" people living in "rich" zip codes, in addition to instructing coprocessor 130 to perform a lookup operation, query execution engine 110 may also have instructed coprocessor 130 (or another coprocessor) to perform a scan operation on the Person table to identify all persons who have an annual salary that is less than $30,000. The result of the scan operation (like the result of the lookup operation) may be in the form of a series of bits (e.g., a bit vector) where each bit corresponds to a different person indicated in the Person table. In one embodiment, query execution engine 110 performs an AND operation on the result of the scan operation and the result of the lookup operation as inputs. Alternatively, coprocessor 130 (or another coprocessor) may be programmed to perform the AND operation. In this embodiment, query execution engine 110 may create another CCB where the operands include a (e.g., virtual) address to the result of the lookup operation and an address to the result of the scan operation.

[0089] As described previously, the size of a lookup vector may not fit entirely in memory of coprocessor 130 at one time. In one of the two scenarios described previously, the lookup vector is divided into four "mini"-vectors and coprocessor 130 operates on each mini-vector separately, thus requiring coprocessor 130 to read in zip code data (from the Person table) for each person four times. The total result produced by coprocessor 130 executing this lookup operation may comprise four separate array of bits, which are eventually OR'd together to yield a single array of bits (again, one for each person indicated in the Person table). This OR'ing step (which may comprise three OR operations) may be performed by query execution engine 110. Alternatively, coprocessor 130 may be programmed to perform the OR operations.

[0090] In the other of the two scenarios, query execution engine 110 causes four different coprocessors to perform a lookup operation using different portions of the lookup vector. Then, the result from one of the coprocessors is OR'd with the result from each of the other coprocessors to yield a single array of bits (one for each person indicated in the Person table). Again, this OR'ing step may be performed by query execution engine 110 or by one of the coprocessors.

[0091] Once query execution engine 110 determines which people live in "rich" zip codes, query execution engine 110 uses that information to determine those people who are also considered "poor," as indicated above.

[0092] While the above description refers to performing either a scan operation or a lookup operation, embodiments may involve one coprocessor performing scan operation for a particular query while another coprocessor is performing a lookup operation for the particular query. Thus, multiple coprocessors may execute simultaneously for the same query but perform different operations.

[0093] An advantage of embodiments described herein is that a general purpose microprocessor may offload data-intensive operations to one or more coprocessors that are separate from the microprocessor in order to free up usage of the microprocessor for other tasks. Thus, the coprocessor(s) may operate asynchronously with respect to the query processing software that causes the coprocessors to perform the operations. Additionally, the one or more coprocessors may perform those operations much faster than the general purpose microprocessor executing the query processing software.

HARDWARE OVERVIEW



[0094] According to one embodiment, the techniques described herein are implemented by one or more special-purpose computing devices. The special-purpose computing devices may be hard-wired to perform the techniques, or may include digital electronic devices such as one or more application-specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs) that are persistently programmed to perform the techniques, or may include one or more general purpose hardware processors programmed to perform the techniques pursuant to program instructions in firmware, memory, other storage, or a combination. Such special-purpose computing devices may also combine custom hard-wired logic, ASICs, or FPGAs with custom programming to accomplish the techniques. The special-purpose computing devices may be desktop computer systems, portable computer systems, handheld devices, networking devices or any other device that incorporates hard-wired and/or program logic to implement the techniques.

[0095] For example, FIG. 5 is a block diagram that illustrates a computer system 500 upon which an embodiment of the invention may be implemented. Computer system 500 includes a bus 502 or other communication mechanism for communicating information, and a hardware processor 504 coupled with bus 502 for processing information. Hardware processor 504 may be, for example, a general purpose microprocessor.

[0096] Computer system 500 also includes a main memory 506, such as a random access memory (RAM) or other dynamic storage device, coupled to bus 502 for storing information and instructions to be executed by processor 504. Main memory 506 also may be used for storing temporary variables or other intermediate information during execution of instructions to be executed by processor 504. Such instructions, when stored in non-transitory storage media accessible to processor 504, render computer system 500 into a special-purpose machine that is customized to perform the operations specified in the instructions.

[0097] Computer system 500 further includes a read only memory (ROM) 508 or other static storage device coupled to bus 502 for storing static information and instructions for processor 504. A storage device 510, such as a magnetic disk, optical disk, or solid-state drive is provided and coupled to bus 502 for storing information and instructions.

[0098] Computer system 500 may be coupled via bus 502 to a display 512, such as a cathode ray tube (CRT), for displaying information to a computer user. An input device 514, including alphanumeric and other keys, is coupled to bus 502 for communicating information and command selections to processor 504. Another type of user input device is cursor control 516, such as a mouse, a trackball, or cursor direction keys for communicating direction information and command selections to processor 504 and for controlling cursor movement on display 512. This input device typically has two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), that allows the device to specify positions in a plane.

[0099] Computer system 500 may implement the techniques described herein using customized hard-wired logic, one or more ASICs or FPGAs, firmware and/or program logic which in combination with the computer system causes or programs computer system 500 to be a special-purpose machine. According to one embodiment, the techniques herein are performed by computer system 500 in response to processor 504 executing one or more sequences of one or more instructions contained in main memory 506. Such instructions may be read into main memory 506 from another storage medium, such as storage device 510. Execution of the sequences of instructions contained in main memory 506 causes processor 504 to perform the process steps described herein. In alternative embodiments, hard-wired circuitry may be used in place of or in combination with software instructions.

[0100] The term "storage media" as used herein refers to any non-transitory media that store data and/or instructions that cause a machine to operate in a specific fashion. Such storage media may comprise non-volatile media and/or volatile media. Non-volatile media includes, for example, optical disks, magnetic disks, or solid-state drives, such as storage device 510. Volatile media includes dynamic memory, such as main memory 506. Common forms of storage media include, for example, a floppy disk, a flexible disk, hard disk, solid-state drive, magnetic tape, or any other magnetic data storage medium, a CD-ROM, any other optical data storage medium, any physical medium with patterns of holes, a RAM, a PROM, and EPROM, a FLASH-EPROM, NVRAM, any other memory chip or cartridge.

[0101] Storage media is distinct from but may be used in conjunction with transmission media. Transmission media participates in transferring information between storage media. For example, transmission media includes coaxial cables, copper wire and fiber optics, including the wires that comprise bus 502. Transmission media can also take the form of acoustic or light waves, such as those generated during radio-wave and infra-red data communications.

[0102] Various forms of media may be involved in carrying one or more sequences of one or more instructions to processor 504 for execution. For example, the instructions may initially be carried on a magnetic disk or solid-state drive of a remote computer. The remote computer can load the instructions into its dynamic memory and send the instructions over a telephone line using a modem. A modem local to computer system 500 can receive the data on the telephone line and use an infra-red transmitter to convert the data to an infra-red signal. An infra-red detector can receive the data carried in the infra-red signal and appropriate circuitry can place the data on bus 502. Bus 502 carries the data to main memory 506, from which processor 504 retrieves and executes the instructions. The instructions received by main memory 506 may optionally be stored on storage device 510 either before or after execution by processor 504.

[0103] Computer system 500 also includes a communication interface 518 coupled to bus 502. Communication interface 518 provides a two-way data communication coupling to a network link 520 that is connected to a local network 522. For example, communication interface 518 may be an integrated services digital network (ISDN) card, cable modem, satellite modem, or a modem to provide a data communication connection to a corresponding type of telephone line. As another example, communication interface 518 may be a local area network (LAN) card to provide a data communication connection to a compatible LAN. Wireless links may also be implemented. In any such implementation, communication interface 518 sends and receives electrical, electromagnetic or optical signals that carry digital data streams representing various types of information.

[0104] Network link 520 typically provides data communication through one or more networks to other data devices. For example, network link 520 may provide a connection through local network 522 to a host computer 524 or to data equipment operated by an Internet Service Provider (ISP) 526. ISP 526 in turn provides data communication services through the world wide packet data communication network now commonly referred to as the "Internet" 528. Local network 522 and Internet 528 both use electrical, electromagnetic or optical signals that carry digital data streams. The signals through the various networks and the signals on network link 520 and through communication interface 518, which carry the digital data to and from computer system 500, are example forms of transmission media.

[0105] Computer system 500 can send messages and receive data, including program code, through the network(s), network link 520 and communication interface 518. In the Internet example, a server 530 might transmit a requested code for an application program through Internet 528, ISP 526, local network 522 and communication interface 518.

[0106] The received code may be executed by processor 504 as it is received, and/or stored in storage device 510, or other non-volatile storage for later execution.

[0107] In the foregoing specification, embodiments of the invention have been described with reference to numerous specific details that may vary from implementation to implementation. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. The sole and exclusive indicator of the scope of the invention, and what is intended by the applicants to be the scope of the invention, is the scope of the set of claims that issue from this application, in the specific form in which such claims issue.


Claims

1. A method for processing a query using a coprocessor (130), comprising:

determining (320) that execution of the query involves a particular operation;

in response to determining that execution of the query involves a particular operation, generating a command (142) that includes, as parameters of the command, first address data that is used to identify input data (144) to be read by the coprocessor, and second address data that is used to identify a lookup vector (400), wherein the lookup vector is generated based on a dimension table;

causing (330) the command to be stored in a memory (140) that is separate from the coprocessor;

processing, by the coprocessor, the command by:

reading (340) the command from the memory;

causing the input data to be read from a location indicated by the first address data, wherein the input data comprises a plurality of values;

causing the lookup vector to be read from a location that is indicated by the second address data;

for each value of the plurality of values, identifying a position within the lookup vector, and retrieving a result from the lookup vector at the position that corresponds to said each value;

generating a result data (146) based on the identifying for each value of the plurality of values;

causing (350) the result data to be stored;
wherein the method is performed by one or more computing devices;
and wherein the steps of determining, generating a command, and causing the command are performed by a general purpose microprocessor (504) executing a query execution engine, wherein the general purpose microprocessor is separate from the coprocessor.


 
2. The method of Claim 1, wherein:

the command further includes operation type data that indicates a type of operation to perform;

the coprocessor processing the command further by first identifying the operation type data to determine the type of operation.


 
3. The method of Claim 2, further comprising:

using the operation type data to determine logic that is used to interpret the input data;

converting the input data from a first data type to a second data type that is different than the first data type.


 
4. The method of Claim 1, wherein:

the input data comprises value data and count data;

an input value in the value data corresponds to data elements, a number of which is indicated by a count value in the count data;

identifying a result within the lookup vector comprises identifying a result that corresponds to the input value;

generating the result data comprises generating a result indication that indicates the result that corresponds to the input value;

generating the result data is performed without identifying, said number of times, a result that corresponds to the input value within the lookup vector.


 
5. The method of Claim 1, wherein the command further includes destination data that indicates where the result data is to be stored.
 
6. The method of Claim 1, wherein causing the result data to be stored comprises causing the result data to be stored in a cache of a microprocessor and/or further comprises causing a completion status to be stored that indicates the command has been performed.
 
7. The method of Claim 1, wherein:

determining that execution of the query involves a plurality of operations that includes the particular operation and one or more other operations;

the method further comprising:

retrieving the result data from storage;

after retrieving the result data from the storage, processing the one or more other operations that require the result data as input to the one or more other operations.


 
8. The method of Claim 1, wherein:

the coprocessor is a first coprocessor of a plurality of coprocessors that are connected in a series;

causing the result data to be stored comprises causing the result data to be sent to a buffer of a second coprocessor of the plurality of coprocessors;

the method further comprising:

reading, by the second coprocessor, the result data from the buffer while the first coprocessor is executing a portion of the query, and

based on the result data, generating, by the second coprocessor, second result data.


 
9. The method of Claim 1, wherein:

generating the command comprises generating a plurality of commands that includes the command;

causing the command to be stored in memory comprises, for each command of the plurality of commands, causing said each command to be stored in the memory;

each coprocessor of a plurality of coprocessors selects a command of the plurality of commands.


 
10. The method of Claim 1, wherein:

the first address data that is included in the command includes one or more virtual addresses;

the method further comprising causing the one or more virtual addresses to be replaced with one or more physical addresses that the coprocessor uses to read the input data.


 
11. The method of Claim 1, further comprising determining whether one or more criteria are satisfied, wherein causing the command to be sent to the coprocessor is only performed if the one or more criteria are satisfied, and optionally wherein the one or more criteria are based on a size of the lookup vector or an amount of data that needs to be read in by the coprocessor to perform the particular operation.
 
12. The method of Claim 1, wherein generating the result data based on the identifying comprises generating a bit vector, wherein each bit in the bit vector indicates whether a value of the plurality of values is found in the lookup vector.
 
13. A coprocessor that is configured to perform the steps of:

receiving a command (1) that was generated by a microprocessor separate from the coprocessor that executes instructions related to query processing and (2) that includes, as parameters of the command, first address data that is used to identify input data to be read by the coprocessor, and second address data that is used to identify a lookup vector, wherein the lookup vector is generated based on a dimension table;

reading the command from a memory that is separate from the coprocessor;

causing the input data to be read from a location indicated by the first address data,

wherein the input data comprises a plurality of values;

causing the lookup vector to be read from a location that is indicated by the second address data;

for each value of the plurality of values, identifying a position within the lookup vector, and retrieving a result from the lookup vector at the position that corresponds to said each value;

generating a result data based on the identifying for each value of the plurality of values;

causing the microprocessor to be informed of the result data.


 
14. The coprocessor of Claim 13, wherein:

the coprocessor is a first coprocessor of a plurality of coprocessors that are connected in a series;

causing the result data to be stored comprises causing the result data to be sent to a buffer of a second coprocessor of the plurality of coprocessors.


 


Ansprüche

1. Verfahren zum Verarbeiten einer Anfrage unter Verwendung eines Coprozessors (130), das Folgendes umfasst:

Bestimmen (320), dass die Ausführung der Anfrage eine spezielle Operation beinhaltet;

in Reaktion auf die Bestimmung, dass die Ausführung der Anfrage eine spezielle Operation beinhaltet, Erzeugen eines Befehls (142), der als Parameter des Befehls erste Adressendaten, die verwendet werden, um durch den Coprozessor zu lesende Eingangsdaten (144) zu identifizieren, und zweite Adressendaten, die verwendet werden, um einen Nachschlagevektor (400) zu identifizieren, umfasst, wobei der Nachschlagevektor auf der Basis einer Dimensionstabelle erzeugt wird;

Bewirken (330), dass der Befehl in einem Arbeitsspeicher (140) gespeichert wird, der vom Coprozessor verschieden ist;

Verarbeiten des Befehls durch den Coprozessor durch:

Lesen (340) des Befehls aus dem Arbeitsspeicher;

Bewirken, dass die Eingangsdaten aus einem Ort gelesen werden, der durch die ersten Adressendaten angegeben wird, wobei die Eingangsdaten mehrere Werte umfassen;

Bewirken, dass der Nachschlagevektor aus einem Ort gelesen wird, der durch die zweiten Adressendaten angegeben wird;

für jeden Wert der mehreren Werte Identifizieren einer Position innerhalb des Nachschlagevektors und Abrufen eines Ergebnisses vom Nachschlagevektor in der Position, die jedem Wert entspricht;

Erzeugen von Ergebnisdaten (146) auf der Basis der Identifikation für jeden Wert der mehreren Werte;

Bewirken (350), dass die Ergebnisdaten gespeichert werden;

wobei das Verfahren durch eine oder mehrere Rechenvorrichtungen durchgeführt wird;

und wobei die Schritte des Bestimmens, Erzeugens eines Befehls und Bewirkens des Befehls, durch einen Universalmikroprozessor (504) durchgeführt werden, der eine Anfrageausführungsmaschine ausführt, wobei der Universalmikroprozessor vom Coprozessor separat ist.


 
2. Verfahren nach Anspruch 1, wobei:

der Befehl ferner Operationstypdaten umfasst, die einen Typ von durchzuführender Operation angeben;

der Coprozessor den Befehl ferner durch zuerst Identifizieren der Operationstypdaten, um den Typ von Operation zu bestimmen, verarbeitet.


 
3. Verfahren nach Anspruch 2, das ferner Folgendes umfasst:

Verwenden der Operationstypdaten, um eine Logik zu bestimmen, die verwendet wird, um die Eingangsdaten zu interpretieren;

Umwandeln der Eingangsdaten von einem ersten Datentyp in einen zweiten Datentyp, der anders als der erste Datentyp ist.


 
4. Verfahren nach Anspruch 1, wobei:

die Eingangsdaten Wertdaten und Zähldaten umfassen;

ein Eingangswert in den Wertdaten Datenelementen entspricht, von denen eine Anzahl durch einen Zählwert in den Zähldaten angegeben wird;

das Identifizieren eines Ergebnisses innerhalb des Nachschlagevektors das Identifizieren eines Ergebnisses, das dem Eingangswert entspricht, umfasst;

das Erzeugen der Ergebnisdaten das Erzeugen einer Ergebnisangabe umfasst, die das Ergebnis angibt, das dem Eingangswert entspricht;

das Erzeugen der Ergebnisdaten ohne Identifizieren eines Ergebnisses in der Anzahl, die dem Eingangswert entspricht, innerhalb des Nachschlagevektors durchgeführt wird.


 
5. Verfahren nach Anspruch 1, wobei der Befehl ferner Zieldaten umfasst, die angeben, wo die Ergebnisdaten gespeichert werden sollen.
 
6. Verfahren nach Anspruch 1, wobei das Bewirken, dass die Ergebnisdaten gespeichert werden, das Bewirken, dass die Ergebnisdaten in einem Cache eines Mikroprozessors gespeichert werden, umfasst und/oder ferner das Bewirken, dass ein Vollendungsstatus gespeichert wird, der angibt, dass der Befehl durchgeführt wurde, umfasst.
 
7. Verfahren nach Anspruch 1, wobei:

Bestimmen, dass die Ausführung der Anfrage mehrere Operationen beinhaltet, die die spezielle Operation und eine oder mehrere andere Operationen umfassen;

wobei das Verfahren ferner Folgendes umfasst:

Abrufen der Ergebnisdaten aus dem Speicher;

nach dem Abrufen der Ergebnisdaten aus dem Speicher Verarbeiten der einen oder der mehreren anderen Operationen, die die Ergebnisdaten als Eingabe in die eine oder die mehreren anderen Operationen erfordern.


 
8. Verfahren nach Anspruch 1, wobei:

der Coprozessor ein erster Coprozessor von mehreren Coprozessoren ist, die in einer Reihe verbunden sind;

das Bewirken, dass die Ergebnisdaten gespeichert werden, das Bewirken, dass die Ergebnisdaten zu einem Puffer eines zweiten Coprozessors der mehreren Coprozessoren gesendet werden, umfasst;

wobei das Verfahren ferner Folgendes umfasst:
Lesen der Ergebnisdaten aus dem Puffer durch den zweiten Coprozessor, während der erste Coprozessor einen Teil der Anfrage ausführt, und

auf der Basis der Ergebnisdaten Erzeugen von zweiten Ergebnisdaten durch den zweiten Coprozessor.


 
9. Verfahren nach Anspruch 1, wobei:

das Erzeugen des Befehls das Erzeugen von mehreren Befehlen umfasst, die den Befehl umfassen;

das Bewirken, dass der Befehl im Arbeitsspeicher gespeichert wird, für jeden Befehl der mehreren Befehle das Bewirken, dass jeder Befehl im Arbeitsspeicher gespeichert wird, umfasst;

jeder Coprozessor von mehreren Coprozessoren einen Befehl der mehreren Befehle auswählt.


 
10. Verfahren nach Anspruch 1, wobei:

die ersten Adressendaten, die im Befehl enthalten sind, eine oder mehrere virtuelle Adressen umfassen;

das Verfahren ferner das Bewirken, dass die eine oder die mehreren virtuellen Adressen durch eine oder mehrere physikalische Adressen ersetzt werden, die der Coprozessor verwendet, um die Eingangsdaten zu lesen, umfasst.


 
11. Verfahren nach Anspruch 1, das ferner das Bestimmen, ob ein oder mehrere Kriterien erfüllt sind, umfasst, wobei das Bewirken, dass der Befehl zum Coprozessor gesendet wird, nur durchgeführt wird, wenn das eine oder die mehreren Kriterien erfüllt sind, und wobei wahlweise das eine oder die mehreren Kriterien auf einer Größe des Nachschlagevektors oder einer Menge an Daten, die durch den Coprozessor eingelesen werden müssen, um die spezielle Operation durchzuführen, basieren.
 
12. Verfahren nach Anspruch 1, wobei das Erzeugen der Ergebnisdaten auf der Basis der Identifikation das Erzeugen eines Bitvektors umfasst, wobei jedes Bit im Bitvektor angibt, ob ein Wert der mehreren Werte im Nachschlagevektor gefunden wird.
 
13. Coprozessor, der dazu konfiguriert ist, die folgenden Schritte durchzuführen:

Empfangen eines Befehls (1), der durch einen vom Coprozessor separaten Mikroprozessor erzeugt wurde, der Anweisungen in Bezug auf eine Anfrageverarbeitung ausführt, und (2), der als Parameter des Befehls erste Adressendaten, die verwendet werden, um durch den Coprozessor zu lesende Eingangsdaten zu identifizieren, und zweite Adressendaten, die verwendet werden, um einen Nachschlagevektor zu identifizieren, umfasst, wobei der Nachschlagevektor auf der Basis einer Dimensionstabelle erzeugt wird;

Lesen des Befehls aus einem Arbeitsspeicher, der vom Coprozessor separat ist;

Bewirken, dass die Eingangsdaten aus einem Ort gelesen werden, der durch die ersten Adressendaten angegeben wird, wobei die Eingangsdaten mehrere Werte umfassen;

Bewirken, dass der Nachschlagevektor aus einem Ort gelesen wird, der durch die zweiten Adressendaten angegeben wird;

für jeden Wert der mehreren Werte Identifizieren einer Position innerhalb des Nachschlagevektors, und Abrufen eines Ergebnisses vom Nachschlagevektor in der Position, die jedem Wert entspricht;

Erzeugen von Ergebnisdaten auf der Basis der Identifikation für jeden Wert der mehreren Werte;

Bewirken, dass der Mikroprozessor über die Ergebnisdaten informiert wird.


 
14. Coprozessor nach Anspruch 13, wobei:

der Coprozessor ein erster Coprozessor von mehreren Coprozessoren ist, die in einer Reihe verbunden sind;

das Bewirken, dass die Ergebnisdaten gespeichert werden, das Bewirken, dass die Ergebnisdaten zu einem Puffer eines zweiten Coprozessors der mehreren Coprozessor gesendet werden, umfasst.


 


Revendications

1. Procédé de traitement d'une requête au moyen d'un coprocesseur (130), comprenant les étapes suivantes :

déterminer (320) que l'exécution de la requête implique une opération particulière ;

en réponse à la détermination que l'exécution de la requête implique une opération particulière, générer une commande (142) qui inclut, en tant que paramètres de la commande, des premières données d'adresse qui sont utilisées pour identifier des données d'entrée (144) à lire par le coprocesseur, et des secondes données d'adresse qui sont utilisées pour identifier un vecteur de consultation (400), où le vecteur de consultation est généré sur la base d'une table de dimensions ;

provoquer (330) le stockage de la commande dans une mémoire (140) qui est séparée du coprocesseur ;

traiter, par le coprocesseur, la commande en :

lisant (340) la commande dans la mémoire ;

provoquant la lecture des données d'entrée depuis un emplacement indiqué par les premières données d'adresse, où les données d'entrée comprennent une pluralité de valeurs ;

provoquant la lecture du vecteur de consultation depuis un emplacement qui est indiqué par les secondes données d'adresse ;

pour chaque valeur de la pluralité de valeurs, en identifiant une position à l'intérieur du vecteur de consultation, et en récupérant un résultat à partir du vecteur de consultation à la position qui correspond à chaque dite valeur ;

générant des données de résultat (146) sur la base de l'identification pour chaque valeur de la pluralité de valeurs ;

provoquant (350) le stockage des données de résultat ;

où le procédé est exécuté par un ou plusieurs dispositifs de calcul ;

et où les étapes comprenant de déterminer, de générer une commande et de provoquer la commande sont exécutées par un microprocesseur à usage général (504) exécutant un moteur d'exécution de requêtes, où le microprocesseur à usage général est séparé du coprocesseur.


 
2. Procédé selon la revendication 1, dans lequel :

la commande comprend en outre des données de type d'opération qui indiquent un type d'opération à exécuter ;

le coprocesseur traite en outre la commande en identifiant d'abord les données de type d'opération pour déterminer le type d'opération.


 
3. Procédé selon la revendication 2, comprenant en outre les étapes suivantes :

utiliser les données de type d'opération pour déterminer la logique qui est utilisée pour interpréter les données d'entrée ;

convertir les données d'entrée d'un premier type en un second type qui est différent du premier type de données.


 
4. Procédé selon la revendication 1, dans lequel :

les données d'entrée comprennent des données de valeur et des données de comptage ;

une valeur d'entrée dans les données de valeur correspond à des éléments de données, dont un nombre est indiqué par une valeur de comptage dans les données de comptage ;

l'identification d'un résultat dans le vecteur de consultation comprend l'identification d'un résultat qui correspond à la valeur d'entrée ;

la génération des données de résultat comprend la génération d'une indication de résultat qui indique le résultat qui correspond à la valeur d'entrée ;

la génération des données de résultat est effectuée sans identifier, ledit nombre de fois, un résultat qui correspond à la valeur d'entrée dans le vecteur de consultation.


 
5. Procédé selon la revendication 1, dans lequel la commande comprend en outre des données de destination qui indiquent où les données de résultat doivent être stockées.
 
6. Procédé selon la revendication 1, dans lequel provoquer le stockage des données de résultat comprend de provoquer le stockage des données de résultat dans un cache d'un microprocesseur et/ou comprend en outre de provoquer le stockage d'un état d'achèvement qui indique que la commande a été exécutée.
 
7. Procédé selon la revendication 1, comprenant l'étape suivante :

déterminer que l'exécution de la requête implique une pluralité d'opérations qui comprend l'opération particulière et une ou plusieurs autres opérations ;

le procédé comprenant en outre les étapes suivantes :

récupérer les données de résultat de la mémoire ;

après la récupération des données de résultat depuis la mémoire, traiter les une ou plusieurs autres opérations qui requièrent les données de résultat comme entrées pour les une ou plusieurs autres opérations.


 
8. Procédé selon la revendication 1, dans lequel :

le coprocesseur est un premier coprocesseur d'une pluralité de coprocesseurs qui sont connectés en série ;

provoquer le stockage des données de résultat comprend de provoquer l'envoi des données de résultat à un tampon d'un deuxième coprocesseur de la pluralité de coprocesseurs ;

le procédé comprenant en outre les étapes suivantes :

lire, par le second coprocesseur, les données de résultat à partir du tampon pendant que le premier coprocesseur exécute une partie de la requête, et

sur la base des données de résultat, générer, par le second coprocesseur, des secondes données de résultat.


 
9. Procédé selon la revendication 1, dans lequel :

générer la commande comprend de générer une pluralité de commandes qui incluent la commande ;

provoquer le stockage de la commande en mémoire comprend, pour chaque commande de la pluralité de commandes, de provoquer le stockage de chaque commande dans la mémoire ;

chaque coprocesseur d'une pluralité de coprocesseurs sélectionne une commande de la pluralité de commandes.


 
10. Procédé selon la revendication 1, dans lequel :

les premières données d'adresse qui sont incluses dans la commande comprennent une ou plusieurs adresses virtuelles ;

le procédé comprenant en outre de provoquer le remplacement des une ou plusieurs adresses virtuelles par une ou plusieurs adresses physiques que le coprocesseur utilise pour lire les données d'entrée.


 
11. Procédé selon la revendication 1, comprenant en outre de déterminer si un ou plusieurs critères sont satisfaits, où provoquer l'envoi de la commande au coprocesseur est exécuté uniquement si les un ou plusieurs critères sont satisfaits, et éventuellement où les un ou plusieurs critères sont basés sur une taille du vecteur de consultation ou une quantité de données qui doit être lue par le coprocesseur pour exécuter l'opération particulière.
 
12. Procédé selon la revendication 1, dans lequel la génération des données de résultat sur la base de l'identification comprend la génération d'un vecteur de bits, où chaque bit dans le vecteur de bits indique si une valeur de la pluralité de valeurs est trouvée dans le vecteur de consultation.
 
13. Coprocesseur qui est configuré pour exécuter les étapes suivantes :

recevoir une commande (1) qui a été générée par un microprocesseur séparé du coprocesseur qui exécute des instructions liées au traitement de requête et (2) qui comprend, comme paramètres de la commande, des premières données d'adresse qui sont utilisées pour identifier des données d'entrée à lire par le coprocesseur, et des secondes données d'adresse qui sont utilisées pour identifier un vecteur de consultation, où le vecteur de consultation est généré sur la base d'une table de dimensions ;

lire la commande dans une mémoire qui est séparée du coprocesseur ;

provoquer la lecture des données d'entrée depuis un emplacement indiqué par les premières données d'adresse, où les données d'entrée comprennent une pluralité de valeurs ;

provoquer la lecture du vecteur de consultation depuis un emplacement qui est indiqué par les secondes données d'adresse ;

pour chaque valeur de la pluralité de valeurs, identifier une position à l'intérieur du vecteur de consultation, et récupérer un résultat à partir du vecteur de consultation à la position qui correspond à chaque dite valeur ;

générer des données de résultat sur la base de l'identification pour chaque valeur de la pluralité de valeurs ;

faire en sorte que le microprocesseur soit informé des données de résultat.


 
14. Coprocesseur selon la revendication 13, dans lequel :

le coprocesseur est un premier coprocesseur d'une pluralité de coprocesseurs qui sont connectés en série ;

provoquer le stockage des données de résultat comprend de provoquer l'envoi des données de résultat à un tampon d'un deuxième coprocesseur de la pluralité de coprocesseurs.


 




Drawing




















Cited references

REFERENCES CITED IN THE DESCRIPTION



This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

Patent documents cited in the description




Non-patent literature cited in the description