(19)
(11)EP 2 161 896 A1

(12)EUROPEAN PATENT APPLICATION

(43)Date of publication:
10.03.2010 Bulletin 2010/10

(21)Application number: 08252966.0

(22)Date of filing:  05.09.2008
(51)International Patent Classification (IPC): 
H04L 29/06(2006.01)
(84)Designated Contracting States:
AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MT NL NO PL PT RO SE SI SK TR
Designated Extension States:
AL BA MK RS

(71)Applicant: Zeus Technology Limited
Cambridge CB4 0WS (GB)

(72)Inventors:
  • Mansell, Ben Ross
    Cambridge CB4 3EU (GB)
  • Midgley, Julian Trowan John
    Waterbeach Cambridgeshire CB35 9LR (GB)
  • Flowerday, Crispin Edward Harold
    Cambridge CB1 2LS (GB)
  • Garrett, Owen John
    Cambridgeshire CB6 3RH (GB)

(74)Representative: Atkinson, Ralph et al
Atkinson & Company Intellectual Property Limited 37-41 Gower Street
London WC1E 6HH
London WC1E 6HH (GB)

  


(54)Supplying data files to requesting stations


(57) Data files are supplied to a plurality of requesting stations (102 to 106) by accessing serving devices (109 to 116). A plurality of requests are received for one or more data files at a traffic management system (101) from requesting stations via an external network (107). A specific server is selected at the traffic management system and a request is issued to the selected server over a local network for the requested file. The requested file is accepted at the traffic management system from the selected server over the local network and the requested file is sent from the traffic management system to the requesting station over external network. The response of selected servers is monitored (302) when responding to issued requests thereby generating monitored responses. The monitored responses are compared (303) against an operational criterion to identify sub-standard operations. Thereafter, where appropriate, measures are taken (305) to reduce the processing demand placed upon the selected server so as to improve the monitored operation of the selected server.




Description


[0001] This application represents the first application for a patent directed towards the invention and the subject matter.

[0002] The present invention relates to traffic managing apparatus for supplying data files to requesting stations. The present invention also relates to a method of supplying data files to a plurality of requesting stations by accessing serving devices.

[0003] Traffic management systems for providing load balancing are known, as described in United Kingdom patent number 2 413 868 and United Kingdom patent number 2 414 136, both in the name of the present applicant. Traffic management systems of this type are deployed when very high demands are placed upon particular websites. Thus a website, in the form of one or more HTML files, may be duplicated across many servers, groups of which may be clustered together with access being provided over a local network. An incoming request for the file is initially sent to the traffic management processor which is then responsible for selecting a particular server, possibly in a round robin fashion, issuing a request to the selected server over the local network and then accepting the requested file, so that it may be in turn sent back to the requesting station.

[0004] When installing a system of this type, it is necessary to make an estimate of the expected level of demand so that web pages will be returned in an acceptable period of time and a requesting station will therefore experience an acceptable grade of service. An acceptable grade of service for example may state that having issued a request to a server, the requested file should start to be received from the server within two seconds. Thus, if 90% of files are returned within this two second period, the grade of service may be considered acceptable. Consequently, the provision of hardware in order to achieve this grade of service should be good enough, given that excessive redundancy would be expensive and ultimately lead to the operation being uncompetitive.

[0005] Given these constraints of providing a level of service that is good enough, problems occur when excessive demand is placed on a particular website. For an information based website, excessive demand may occur when a particular event takes place. Similarly, for a commercial website excessive demand may occur when special offers are available or as a result of a successful advertising campaign. Under these circumstances, an unexpected level of demand may be placed on the website resulting in a substandard level of service which may result in customers waiting too long or, ultimately, in requests being lost. Such an experience may have detrimental side effects in that customers may be reluctant to access the website again and these would include regular customers.

[0006] According to a first aspect of the present invention, there is provided a traffic managing apparatus of the aforesaid type, as set out in claim 1.

[0007] According to a second aspect of the present invention, there is provided a method of supplying data files to a plurality of requesting stations by accessing serving devices, of the aforesaid type, as set out in claim 7.

[0008] The invention will now be described by way of example only with reference to the accompnaying drawings, of which:

Figure 1 shows data distribution over the Internet, including a traffic managing system;

Figure 2 details the traffic managing system identified in Figure 1;

Figure 3 shows procedures performed by a processing device identified in Figure 2;

Figure 4 illustrates a protocol diagram for components identified within Figure 3;

Figure 5 details procedures for monitoring server response, identified in Figure 3;

Figure 6 details procedures for comparing responses, identified in Figure 3;

Figure 7 illustrates the taking of measures, of the type identified in Figure 3; and

Figure 8 illustrates server load.


Figure 1



[0009] Traffic managing apparatus 101 is illustrated in Figure 1 that is configured to supply data files to requesting stations 102, 103, 104, 105, 106 etc via the Internet 107. Physical serving devices 109, 110, 111, 112, 113, 114, 115 and 116 may be organised into server pools configured to provide different functionality. However, the present invention does not need to take account of the capabilities of each server and other procedures will be concerned with directing appropriate requests. The present invention is concerned with monitoring the response of the servers and taking appropriate action so as not to overload the servers.

[0010] In order to facilitate the configuration of the servers and the traffic management system 101, an administration station 121 is provided, having human interface peripheral equipment such as a monitor and keyboard etc. In addition, a data carrying medium such as a DVD 122 is illustrated as a means for supplying executable instructions for the traffic managing process. Alternatively, traffic managing instructions may be loaded onto the traffic management system from an external source and the transfer of such instructions may be controlled via the administration system 121. The servers 109, the administration station 121 and the traffic manager 101 communicate via a high bandwidth internal network 108.

Figure 2



[0011] The traffic managing system 101 is detailed in Figure 2 and comprises a first interface 201 to the external Internet 107 for receiving requests from the requesting stations 102 to 106 and for sending a requested file back to the requesting station. In addition, there is provided a second interface 202 for interfacing to internal network 108 and this in turn issues requests to the servers 109 to 116 to effect load balancing and for receiving requested files from a requested server. The traffic managing system 101 also includes a processing device 203 and said devices 201, 202 and 203 communicate via an internal bus 204.

Figure 3



[0012] Procedures performed by the processing device 203 of Figure 2 are illustrated in Figure 3. At step 301 a request is received from a client and at step 302 a question is asked as to whether substandard operation is taking place. When answered in the affirmative, the URL is changed at step 303. Thereafter, at step 304 communication with station servers takes place. Thus, if the question asked at step 302 is answered in the negative, step 303 is bypassed and server communication takes place at step 304.

[0013] At step 305 the server response is monitored and at step 306 the monitored response is compared with the previous responses and where appropriate a substandard operation flag is set. Thus, when set in this way, as a result of monitoring substandard response, the question asked at step 302 will be answered in the affirmative.

Figure 4



[0014] A protocol diagram is illustrated in Figure 4 detailing procedures 301 for the communications between the traffic management system 101 and the requesting stations 102 to 106 and between a traffic management system 101 and the servers 109 to 116.

[0015] For the purposes of illustration, it is assumed that the load balancing activities of the traffic management system 101 will effectively adopt a round robin approach. Thus, traffic manager 120 receives a request 401 in the form of a URL from the Internet 107. In normal operation, this URL is conveyed 402 to the server 109. At server 109 appropriate action is taken in order to generate the requested page which is then returned as HTML 403. The returned HTML is accepted at the traffic management system 120 and then conveyed as 404 back to the requesting station, via the Internet 101.

[0016] The next request is shown being received at 405. On this occasion, the request is issued as shown at 406 to the next server 110. Again, server 110 takes measures in order to generate the requested file and then the generated file is accepted back at the traffic management system 120 as illustrated at 407. Thereafter, the file is sent to the requesting station via the Internet 101 as illustrated at 408.

[0017] Upon receiving the next request, as illustrated at 409, the traffic manager 120 issues a request, as illustrated at 410, to the next server 111. Again, server 111 takes the necessary action required in order to generate the requested data which is then accepted by the traffic management system 120 as illustrated at 411. Thereafter, the accepted file is sent to the requesting station via the Internet 101, as illustrated at 412.

[0018] It is generally accepted practice for the servers 109 to 116 themselves to be multitasking and with each request issued by the traffic management system 101 a new thread will be instantiated. Furthermore, the processing capability available at the servers will be distributed so as to satisfy as many requests in parallel as possible. However, a physical limit will occur, particularly when there is a high demand placed on each of the servers and particularly when the amount of processing required to generate each page is relatively large. Thus, although the load balancing exercises performed by the traffic management system 101 will endeavour to distribute the load upon the servers and possibly take action should part of the network fail, ultimately a point will be reached at which it is not possible for all of the requests to be satisfied to the desired level of service. As this point is reached, it would be possible for manual intervention to take place so as to modify the way in which the servers respond to the requests. However, such an approach may be considered as a brutal intervention in that, having made the modification, further manual intervention will be required in order to return the system back to its preferred mode of operation. Consequently, it is possible that demand could diminish and an unnecessary intervention could remain in place. Thus, the present invention seeks to monitor server response, compare the response against an established criterion, determine whether substandard operation is taking place and then, where appropriate, take measures in order to reestablish standard operation in terms of grade of service.

Figure 5



[0019] As previously described with respect to Figure 3, process 302 monitors the server response so that appropriate measures may be taken should this be considered appropriate. In order to obtain an indication of the operational condition of each of the servers, it would be possible to look at the total number of connections that are concurrently running or alternatively it would be possible to look at the number of connections to a particular node connected to a handler request. Another alternative approach would be to execute code at each of the servers which would then in turn provide an indication back to the traffic management system 101 as to how loaded they actually are. Thus, using a mechanism of this type, it would be possible to obtain a direct indication of the extent to which the processing capabilities within each of the servers is being used. A further alternative approach would be to consider the load placed on the processing device 203 at the traffic management system 101 itself. However, a preferred approach for monitoring server response is detailed in Figure 5.

[0020] Essentially, the preferred approach for monitoring server response consists of measuring the time taken between the traffic manager 101 receiving a request from the Internet 107 and the point at which the requested file begins to be accepted by the traffic management system. This period represents the amount of time required in order for a server to receive a request and perform the necessary processing operations in order to start serving the requested file. In addition, this period also includes any preprocessing performed at the processing device 203, within the traffic management system 101.

[0021] At step 501 the fact that a request has been received at the traffic management system 101 is detected. Thereafter, at step 502 a monitoring clock starts to run. At step 503 this particular process waits while the request is processed and then issued at step 504. The waiting stage is illustrated at step 505.

[0022] At step 506 the start of the acceptance of the data file from the selected server is detected such that at step 507 the clock is stopped and the time is then recorded at step 508.

[0023] Thus, a preferred procedure (illustrated in Figure 5) calculates the time taken from a request being received at the traffic management system 101 and the start of the requested file being accepted back at the traffic management system 101. It is then necessary for a comparison to take place as illustrated at step 303 in order that a question may be asked at step 304 as to whether substandard operations taking place.

[0024] It is therefore necessary to have a criterion or criteria establishing what is considered to be an acceptable grade of service. For the purposes of this illustration, it will be assumed that the level of service being provided is considered to be acceptable if 80% of requests are handled within two seconds. Thus, a single instance of a request taking longer than two seconds would not be considered sufficient to result in the question asked at step 304 being answered in the affirmative, to the effect that the operation is considered to be substandard. Substandard operation occurs when 20% of requests are not processed within the specified two second period. It can be appreciated that those skilled in the art could develop many algorithms in order to determine whether the criteria are met such that the question posed at step 304 may be implemented.

Figure 6



[0025] The following, described with reference to Figure 6, provides an example for illustration purposes only and it should be appreciated that many algorithms of this type would satisfy the requirements of an enabling embodiment.

[0026] At step 601, the time recorded at step 508 is read and in this example identified as time T0. At step 602 an acceptable time is read, ie the maximum acceptable time which in this example has been identified as two seconds.

[0027] At step 603 a question is asked as to whether T0, the measured time, is bigger than the acceptable time TA. If T0 is larger than TA the access time as previously described and as recorded as T0, will be greater than two seconds. Under these circumstances service level zero (SLO) is recorded as being equal to one at step 604. If the question asked at step 603 is answered in the negative, to the effect that T0 is smaller than or equal to TA, the else condition at step 605 results in SLO being set to zero.

[0028] As previously stated, a single instance does not allow a decision to be made and is necessary for an averaging process to be performed over a number of samples. Thus, in this embodiment, the value for the service level is added to a first in first out (FIFO) buffer 606 which allows ten values to be added together. Thus, the ones and zeros generated by procedures 603 to 605 are added to FIFO 606 resulting in the previous ten samples being included within the FIFO. The contents of the FIFO are therefore added at step 607.

[0029] An unacceptable result will cause a one to be generated and an acceptable result will cause a zero to be generated. Unacceptable operation has been established at the 20% value, therefore if FIFO 606 contains two or more values of one, the total when added together will be a value greater than one, and this is considered to be unacceptable operation. Thus, in this simplistic embodiment, a single instance of a long access time in ten samples would be disregarded. However, should two or more such occurrences take place within the ten sample period, the question asked at step 304' (being equivalent to step 304) results in measures being taken at step 305' (equivalent to step 305).

[0030] In a preferred embodiment, the sample period may take place over a much longer duration and the algorithm performed may be more sophisticated.

[0031] In an alternative embodiment it may be possible to provide a more refined degree of optimisation so that measures are only taken when absolutely necessary and the maximum number of transactions take place without these additional measures being enforced. Thus, based on an assessment of the server response, it could be possible for the measures to be taken for a proportion of the transactions, in which this proportion gradually increases as the server response continues to degrade. Thus, as degradation is identified the procedures may be such that only 10% of transactions involve measures being taken. However, if this is not sufficient to relieve the burden placed upon the servers, measures may be taken for 20% of the transactions and then 30% of the transactions and so on until measures are taken for all of the transactions.

[0032] In an alternative embodiment it may be more appropriate to prevent a situation in which a flip flopping occurs between measures being taken and measures not being taken. Thus, for a particular application the switching from a state in which measures are to being taken to a state in which measures are taken and then switching back to the measures not being taken again may place an unnecessary burden upon equipment. For example, it may be necessary for a different area of disc to be accessed thereby resulting in regular large movements of disc heads which in turn would place strain upon the servers and ultimately reduce their life expectancy. Consequently, in some environments it may be preferable to introduce a degree of hysteresis. Thus, in accordance with this preferred embodiment, a preferred system may be established such that it appears reluctant to force measures to be taken in order to relieve strain upon the servers. Thus, a greater level of degradation may be considered acceptable before measures are taken. Thus, when in its first state of normal operation, a criteria may be established such that 80% of transactions take no longer than three seconds to be serviced. Thus, only when these criteria are not being met will the operation be considered to be substandard and the question asked at step 304 will be answered in the affirmative, resulting in measures being taken. However, now that the question asked at step 304 has been answered in the affirmative measures will continue to be taken until a much higher level of performance is achieved. Thus, for the purposes of illustration, standard operation may be specified as 90% of all transactions taking no more than one second. Thus, once measures start to be taken, they will tend to be continued until demand has decreased significantly and only then will operation return to its preferred state.

[0033] Thus, it can be appreciated that many modifications are available in terms of making the assessment as to whether satisfactory operation is taking place and a preferred approach may be constrained by other limitations within the system as a whole.

Figure 7



[0034] An illustration of measures that may be taken upon identifying substandard operation as illustrated in Figure 7. In this example, an original file includes a high definition image 701 and the serving of this high definition image will incur a relatively large demand upon the server concerned. When the level of operation is identified as being substandard, measures are taken such that instead of serving high definition image 701, a substantially similar image 702 of lower definition is served as a replacement. Thus, a web page could include several images of this type or a web page could be totally reconfigured such that substantially the same information is contained therein but the actual amount of data required to transmit the information has been substantially reduced.

[0035] Thus, an example of measures 305 are illustrated in Figure 7. Thus, at step 703 the URL is identified and at step 704 the URL identified at step 703 is replaced with an alternative URL. Thus, the alternative URL points to an alternative file, which substantially contains the same information but in such a way that requires less data and therefore has a shorter access time. Thus, in this specific example, the low bandwidth URL replacement is selected such that picture 702 is served as an alternative to high definition picture 701.

[0036] In an alternative configuration, a modification of image quality could occur without selecting an alternative URL but by performing alternative measures at the server itself. Thus, the data file could be read in a different way such that a decimation of image data occurs.

[0037] In an alternative embodiment, it may be necessary for a server to perform calculations in order to generate a requested web page. For example, the server may be configured to generate an image showing a map in response to a location being specified. However, it may be possible for the degree of detail contained within the map to be modified depending upon a detection of substandard operation. Thus, in a preferred mode of operation a map may contain considerable detail showing images of particular buildings for example. However, when downgraded, the images of the buildings could be replaced by symbols and the degree of detail contained within the map could be reduced significantly. However, the essential information content remains the same.

[0038] It is also appreciated that files other than web pages may be used within this environment. Thus, the stored files may represent video clips and as such video clips of differing definition may be supported and selected depending upon the identification of substandard operation. In an alternative embodiment the service itself may include web pages or similar files that are updated on a regular basis. Consequently, in the preferred mode of operation these files are generated each time a request is made. However, when substandard operation starts to occur, it may be preferable to cache files such that the same data may be served several times with updates occurring on a much less frequent basis. Thus, it should be appreciated to those skilled in the art that many other types of degradation may occur and be considered appropriate for the measures taken at step 305 without departing form the essential nature of the invention.

Figure 8



[0039] A graph for illustrative purposes only is illustrated in Figure 8 in which server load has been plotted against the number of requests made. This shows operation of the server within the environment of the present invention effectively taking place in three possible modes. In a first mode of operation 801 it is not necessary for any measures to be taken given that the load placed on the servers is relatively low and all of the requested data is being supplied back at full bandwidth. Thus, for example requests may be received at the traffic manager at a rate of say five per second. At the other extreme, the traffic manager may be receiving requests in the order of thousands per second resulting in the third mode of operation 803 where the enforcement of measures being taken occurs on a permanent basis. Thus, so many requests are being received that it is not possible for the servers to operate in their preferred mode of operation without introducing substantial delays and occasionally dropping connections. Thus, measures are taken at step 305 for all transactions which may then result in a degradation of spatial definition for all data sent back to the requesting stations.

[0040] The intermediate mode of operation identified as 802 may be considered as that where the measures taking place are having substantial effect upon the quality of service provided by the system. In the example of Figure 8 there is no hysteresis such that having implemented the taking of measures the demand placed upon the servers may be considered as relatively constant as the number of requests increases. However, eventually, the number of requests will be such that the server load starts to increase substantially as the third mode of operation 803 is entered.

[0041] Thus, to summarise, there is a first mode of operation in which it is never necessary to take measures. There is then a second mode of operation during which measures are taken on some occasions and then not on others. Thus, measures may be taken for a proportion of the transactions or a flip flopping may occur between normal operation and modified operation. Finally, there is a third mode of operation in which measures are taken all of the time and all transactions take place in what may be considered a substandard mode of operation.

[0042] The embodiment monitors server response to determine a degree of load placed upon the servers. Secondly, having measured the response, a comparison takes place to determine whether the system is operating as desired or whether substandard operation is taking place. Finally, when the substandard mode of operation is identified, measures are taken in order to rectify this such that requesting stations still receive web pages within an acceptable period of time.


Claims

1. Traffic managing apparatus for supplying data files to a plurality of requesting stations, by accessing serving devices, comprising:

a first interface to an external network for receiving requests from a plurality of requesting stations and for sending a requested file back to a requesting station;

a second interface to an internal network for issuing requests to a plurality of servers to effect load balancing and for receiving a requested file from a requested server; and

a processing device configured to monitor the response of the servers, characterised by being configured to:

monitor the response of selected servers when responding to issued requests, thereby generating monitored responses;

compare said monitored responses against an operational criterion to identify sub-standard operation; and

take measures to reduce the processing demand placed upon the selected server so as to improve the monitored operation of said selected server.


 
2. A method of supplying data files to a plurality of requesting stations by accessing serving devices, comprising the steps of:

receiving a plurality of requests for one or more data files at a traffic management system from a plurality of requesting stations via an external network;

selecting a specific server (the selected server) at said traffic management system;

issuing a request to said selected server over a local network for the requested file;

accepting the requested file at the traffic management system from said selected server over said local network; and

sending the requested file from the traffic management system to the requesting station over the said external network, characterised by:

monitoring the response of selected servers when responding to issued requests, thereby generating monitored responses;

comparing said monitored responses against an operational criterion to identify sub-standard operation; and

taking measures in response to said comparison to reduce the processing demand placed upon the selected server so as to improve the monitored operation of said selected server.


 
3. The method as claimed in claim 2, wherein said data files are HTML web pages or video files.
 
4. The method as claimed in claim 2, wherein copies of the same data file may be obtained from each of a plurality of servers and the traffic management system selects one of said servers in order to perform a traffic balancing operation.
 
5. The method as claimed in claim 2, wherein said measures to reduce processing demand are taken when load balancing has been substantially optimised.
 
6. The method as claimed in claim 2, wherein the response measured is the time taken between the traffic management system starting to process a request from a client and starting to send a response to the client.
 
7. The method as claimed in claim 2, wherein the response measured is the processor demand being placed upon the selected server or the processing demand being placed upon the traffic management system.
 
8. The method as claimed in claim 2, wherein the response is compared against a first criterion to invoke said taking of measures, whereafter the response is measured against a second criterion for the return to normal operation.
 
9. The method as claimed in claim 2, wherein the number of issued requests that are affected by the measures taken is proportionate to the degree of sub-standard operation as determined by said comparing step.
 
10. The method as claimed in claim 2, wherein said criterion is defined in terms of a proportion of the requests issued to the selected server resulting in the file being sent within a specified time.
 
11. The method as claimed in claim 2, wherein said measures involve replacing details of the originally received request with alternative details for the request sent to the selected server.
 
12. The method as claimed in claim 11, wherein the data file is identified by a universal resource locator (URL) and a received URL is replaced with an alternative sent URL.
 
13. The method as claimed in claim 12, wherein the sent URL identifies a substantially similar file to the received URL with similar information but containing less data so as to improve the response of the selected server.
 
14. The method as claimed in claim 11, wherein the file identified by the sent URL has lower resolution images linked thereto compared to the file identified by the received URL.
 
15. The method as claimed in claim 14, where the file identified by the sent URL requires less computational processing to be performed at the server compared to the file identified by the received URL.
 
16. The method as claimed in claim 12, wherein a response to a URL is sent from the traffic management system using a cached copy as an alternative to contacting a server to request data.
 
17. The method as claimed in claim 2, wherein said measures involved sending the requested file directly from the traffic management system, without sending the request to the server, having stored a copy of the requested file on the traffic management system when handling a previous request.
 




Drawing




























Search report










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