(19)
(11)EP 0 483 455 A1

(12)EUROPEAN PATENT APPLICATION

(43)Date of publication:
06.05.1992 Bulletin 1992/19

(21)Application number: 91109957.0

(22)Date of filing:  18.06.1991
(51)International Patent Classification (IPC)5G06F 13/374
(84)Designated Contracting States:
DE FR GB

(30)Priority: 29.10.1990 US 605287

(71)Applicant: International Business Machines Corporation
Armonk, N.Y. 10504 (US)

(72)Inventor:
  • Kalajainen, Dennis M.
    Manassas, VA 22110 (US)

(74)Representative: Jost, Ottokarl, Dipl.-Ing. 
IBM Deutschland Informationssysteme GmbH, Patentwesen und Urheberrecht
D-70548 Stuttgart
D-70548 Stuttgart (DE)


(56)References cited: : 
  
      


    (54)Single request arbitration scheme for common bus


    (57) The invention discloses a single request arbitration code for determining if only one communication module (14) is contending for a common data bus (12) and gives that module (14) access within two bus clock cycles. An aggregate code is generated by using the true and complement of the module identification. A check code for each bit is determined by adding the aggregate code for the bit and its adjacent bit. If the check code contains any zeros, then more than one module (14) is contending for the bus (12). The 10-bit single arbitration scheme allows for error detection and correction on a 32-bit common data bus (12).




    Description

    Background of the Invention


    1. Technical Field



    [0001] The invention disclosed broadly relates to bus communication systems and more particularly relates to the arbitration of control of the bus by contending devices desiring access to a common bus for the transfer of data.

    2. Background Art



    [0002] In high speed communication systems having a common bus interconnected to multiple communicating modules, latency due to arbitration of contending modules and message transfer is a serious problem. When a plurality of communicating modules vie for a common data bus simultaneously, an arbitration code is necessary to prioritize the requesting modules to select the module to receive access to the bus.

    [0003] A bus acquisition system is disclosed in US-A-4,736,366 (Rickard) and a bus transceiver is disclosed in US-A-4,756,006 (Rickard), both patents owned by the common assignee of this invention. The teachings of these two patents are herein incorporated by reference.

    [0004] A typical communication system may have up to 32 communicating modules which compete for a common bus.

    [0005] One hundred twenty-eight logical priority levels can be ascribed to the 32 modules. The arbitration sequence, which takes place upon the data bus, requires eight bus cycles to resolve the winner's logical and physical ID. There are times when only one module is in contention for the bus. Even though it is common for only one module to be vying for the bus at any given time, that module must suffer a full eight-cycle latency of arbitration sequence.

    [0006] A better technique to achieve low latency arbitration would be to add two cycles at the beginning of the arbitration sequence to determine if one or more modules are contending for control of the bus. If only one module is contending for the bus, it is not necessary to continue to the eight-cycle arbitration sequence, but give that module immediate access to the bus.

    Objects of the Invention



    [0007] It is therefore an object of the invention to provide an improved bus arbitration system which reduces the arbitration cycle latency.

    [0008] It is another object of the invention to provide a bus arbitration scheme having error detection.

    [0009] It is still another object of the invention to provide an improved bus arbitration code which can be duplicated or triplicated on a 32-bit data bus to allow error detection or correction, respectively.

    Summary of the Invention



    [0010] The present invention is a method for determining arbitration within a common bus communication system, wherein the module ID is written in true and complement form. All modules determine the number of contending modules by calculating a check code after receiving an aggregate (wire-OR) idle bus code from the bus. If the number of contenders is equal to one, then the winner of the vie has been determined and the arbitration is complete. If the number of contenders is greater than one, then the vie sequence continues to determine the eventual winner.

    Brief Description of the Drawings



    [0011] The present invention, together with other and further advantages and features thereof, will be readily understood from the following detailed description when taken with the accompanying drawings.
    Fig. 1
    is a schematic representation of a common bus having 32 modules connected thereto.
    Fig. 2
    is a timing diagram of the bus clock having a single competitor arbitration code resolution.
    Fig. 3
    is a timing diagram for the bus having multiple module contenders.
    Fig. 4
    is a block diagram showing an implementation of the idle bus arbitration having error correction.

    Detailed Description of the Invention



    [0012] Shown in Fig. 1, a communications unit 10 employs a common data bus 12 and has a plurality of communication modules 14. Each communication module 14 has a physical ID from zero to 31 and contains a bus interface unit (BIU) 16, a central processing unit (CPU) 18 and a memory unit 20. The BIU 16 does the interfacing between the bus 12 and the communication module 14. The central processor 18 controls the communication module, directing the storage of data in memory unit 20 and sending it out on the common bus through the BIU 16.

    [0013] When more than one module requests the bus 12, an arbitration code is employed to decide which contender vying for the bus 12 receives control of the bus 12. A standard method for achieving this would be to force each contending module 14 to output a code of zero through 31 corresponding to its unique 5-bit physical ID. Five bits are needed to represent a number from zero to 31. The codes for each module are wire-OR'ed on the bus (low active) resulting in an aggregate arbitration code. Then the 32-bit aggregate code could be checked for the number of bits that are active. If only one bit is active, bus tenure can be transferred to the corresponding module 14. If greater than one bit is active, then normal arbitration techniques would be employed. However, checking this code for the number of active bits requires a large amount of logic. The logic is usually open-collector bipolar having large drive and sink current requirements. A large number of comparators are needed. Since the entire 32-bit data bus is required to source the code, error detection is difficult, if not impossible.

    [0014] A better code for each contender would be as follows:
    for each logic 0 in the physical ID, output '01' on the bus, and
    for each logic 1 in the physical ID, output '10' on the bus.

    [0015] This forms a true and complement of the physical module ID. Non-contenders would activate no lines on the bus. The result for a five-bit physical ID is a 10-bit single competitor arbitration code. The arbitration code is as follows:



    [0016] The aggregate arbitration code is created by the wire-OR of the bus. For example, two modules 14 are contending, module 0 has a module ID of MID=00000 and module 1 has MID=00001. The resulting aggregate code is 0101010100 as shown in Table 2.



    [0017] A check code can be derived from the aggregate arbitration code satisfying the following equations:



    [0018] If the check code contains any zeros, then more than one module 14 is contending for the bus. Normal arbitration sequences will be required to determine the bus control. If the check code contains all ones, then the bus control can be transferred to the single competing module 14. Table 3 shows the result of a two request arbitration. Table 4 shows the result of a single request arbitration. Note that the check code is equal to all ones in Table 4.





    [0019] This technique allows assignment of tenure in two cycles instead of eight cycles. The logic to check this code compared with the 1-of-32 code approach is significantly simpler and requires much less chip area and wiring bays. Shown in Fig. 2 is a clock step sequence for the two-cycle arbitration code. In step A, all contending modules assert their idle bus code of true and complement of their physical location ID. In step B, all modules determine the number of contending modules by calculating the check code after receiving the aggregate wire-OR idle bus code from the bus. If the number of contenders is equal to one, then the winner of the vie has been determined and the vie is complete. Therefore, the bus state in step C is HO which stands for heading zero of the first message. The winner begins to transfer its message.

    [0020] In Fig. 3 there is shown an arbitration vie resulting in more than one contender. In step C, a bus state is V1 (vie 1). The vie sequence continues to determine the eventual winner. Six more cycles are required.

    [0021] Another feature of the invention is that since the code is compact, it can be duplicated or triplicated on a 32-bit data bus to allow error detection or correction, respectively. That is, only 10 bits are required for the single request arbitration code. In a 32-bit data bus, the arbitration code can be replicated three times. This is not possible with the 1-of-32 code approach since all available data bus lines are used to source the code. An additional error detection can be achieved by the generation of an error code. The error code equations are:



    [0022] If the error code contains any ones, then it can be assumed that an error has occurred. This is true since each pair of aggregate code bits should never contain the '11' combination. The example shown in Table 5 shows the error code.



    [0023] Error code correction can be done by a voting of two out of three of the triplicated idle bus arbitration logical implementation. Shown in Fig. 4 are the 10-bit arbitration codes put in 10 bits 0-9. In Fig. 4, the three bits for bit 9 are first AND'ed together, then OR'ed together to form the triple modular redundant unit 22, which are AND'ed together to form the bit error correction code. These significant bits are OR'ed together to produce the error code which must equal zero if no error has occurred.

    [0024] The single request arbitration code has applications to many different bus protocols. It can be implemented to provide extensions and enhancements to the current common bus protocols.


    Claims

    1. A method of single request arbitration between contending communication modules (14) connected to a common data bus (12) comprising the steps of:

    generating an aggregate code by forming the true and complement binary numbers of a contending module (14) identification;

    asserting during a first clock cycle the aggregate code to the bus (12);

    calculating a check code from the asserted aggregate code; and

    determining during a second clock cycle if the number of contending modules (14) is one.
     
    2. The method of claim 1, wherein the aggregate code is generated in duplicate to provide error detection.
     
    3. The method of claim 1, wherein the aggregate code is generated in triplicate to provide error correction.
     
    4. The method of claim 1, wherein an error code is produced by the logical ANDing of the aggregate code.
     
    5. A communication system having a common data bus (12) employing a single request arbitration code among contending communication modules (14) comprising:

    means for generating an aggregate code comprising the true and complement binary numbers of a contending module (14) identification;

    means for asserting the aggregate code on the bus (12);

    means for calculating a check code by logical ORing the aggregate code; and

    means for determining if only one module (14) is contending for the bus (12) based on the clock code;

    whereby access to the bus (12) is given to the sole contending module (14).
     
    6. The system of claim 5, wherein the generating means produces the aggregate code in duplicate to provide error detection.
     
    7. The system of claim 5, wherein the generating means produces the aggregate code in triplicate to provide error correction.
     
    8. The system of claim 5, wherein an error code is generated by the logical ANDing of the aggregate code.
     




    Drawing













    Search report