WebSphere MQ for Windows V6.0 - Performance Evaluations. Version 1.1. August Peter Toghill

Transcription

1 WebSphere MQ for Windows V6.0 - Performance Evaluations Version 1.1 August 2005 Peter Toghill WebSphere MQ Performance IBM UK Laboratories Hursley Park Winchester Hampshire SO21 2JN Property of IBM

2 Take Note! Before using this report, be sure to read the general information under Notices. First Edition, June 2005 This edition applies to V1.0 of WebSphere MQ for Windows V6.0 Performance Evaluations and to all subsequent releases and modifications until otherwise indicated in new editions. (C) Copyright International Business Machines Corporation All rights reserved. Note to U.S. Government users Documentation related to restricted rights Use, duplication or disclosure is subject to restrictions set forth in GSA ADP Schedule contract with IBM corp. Page II

3 Notices This report is intended to help the reader understand the performance characteristics of WebSphere MQ for Windows 2003 V6.0. The information is not intended as the specification of any programming interfaces that are provided by WebSphere MQ. References in this report to IBM products or programs do not imply that IBM intends to make these available in all countries in which it operates. Information contained in this report has not been submitted to any formal IBM test and is distributed as-is. The use of this information and the implementation of any of the techniques is the responsibility of the customer. Much depends on the ability of the reader to evaluate the information and project the results to their operational environment. The performance measurements included in this report were measured in a controlled environment and the results obtained in other environments may vary significantly. Trademarks and service marks: The following terms used in this publication are trademarks of the IBM Corporation in the United States or other countries or both: IBM MQSeries WebSphere MQ SupportPac FFST AIX UNIX is a registered trademark and licensed exclusively through X/Open Company Limited. Windows is a registered trademark and licensed exclusively through Microsoft Microsoft, Windows, and Windows 2003 are trademarks of Microsoft Corporation in the United States, other countries, or both.

4 Preface Target audience This SupportPac is designed for people who: Will be designing and implementing solutions using WebSphere MQ for Windows. Want to understand the performance limits of WebSphere MQ for Windows V6.0. Want to understand what actions may be taken to tune WebSphere MQ for Windows. The reader should have a general awareness of the Windows 2003 operating system and of MQSeries in order to make best use of this SupportPac. Readers should read the section How this document is arranged Page VI to familiarise themselves with where specific information can be found for later reference. The contents of this SupportPac This SupportPac includes: Release highlights performance charts. Performance measurements with figures and tables to present the performance capabilities of WebSphere MQ local queue manager, client channel, and distributed queuing scenarios. Interpretation of the results and implications on designing or sizing of the WebSphere MQ local queue manager, client channel, and distributed queuing configurations. Feedback on this SupportPac We welcome constructive feedback on this report. Does it provide the sort of information you want? Do you feel something important is missing? Is there too much technical detail, or not enough? Could the material be presented in a more useful manner? Please direct any comments of this nature to WMQPG@uk.ibm.com. Specific queries about performance problems on your WebSphere MQ system should be directed to your local IBM Representative or Support Centre. Acknowledgements The author is very grateful to Richard Eures for his significant contribution to V1.0 of this report. Page IV

5 Introduction The three scenarios used in this report to generate the performance data are: Local queue manager scenario. Client channel scenario. Distributed queuing scenario. Unless otherwise specified, the standard message sized used for all the measurements in this report is 2K (2,048 bytes). An IBM x-series 365 machine (4-way 3 GHz, 4 GB RAM) running Windows 2003 was used as the device under test for all the measurements in this report. Page V

6 How this document is arranged Performance Headlines Pages: 1-13 Section one contains the performance headlines for each of the three scenarios, with MQI applications connected to: A local queue manager. A remote queue manager over MQI-client channels. A local queue manager, driving throughput between the local and remote queue manager over server channel pairs. The headline tests show: The maximum message throughput achieved with an increasing number of MQI applications. The maximum number of MQI-clients connected to a queue manager. The maximum number of server channel pairs between two queue managers, for a fixed think time between messages until the response time exceeds one second. Large Messages Pages: Section two contains performance measurements for large messages. This includes MQI response times of 50byte to 2MB messages. It also includes 20K, 200k and 2M messages using the same scenarios as for the Performance Headlines. Application Bindings Page: Section three contains performance measurements for 'trusted, normal, and isolated' server applications, using the same three scenarios as for the Performance Headlines. Short & Long Sessions Page: 46 This section was introduced in Version V1.1 of this document. Section four contains performance measurements for short sessions. That is, an MQI application connecting to the queue manager, processing a few messages between connecting to and disconnecting from the queue manager. Error! Reference source not found. Pages: Error! Bookmark not defined. This section was introduced in Versoin V1.1 of this document. Section five of this document shows: The number of MQI-client channels that were connected into a single queue manager, with a server application processing one nonpersistent round trip per MQI-client per minute. The number of server channel pairs that were connected between two queue managers on separate server machines, with a server application processing one nonpersistent round trip per server channel pair per minute. Tuning Recommendations Pages: In previous SupportPacs, tuning recommendations have been in a separate section. In this document, queue manager parameters are mentioned with the measurements they are appropriate to. Page VI

7 Measurement Environment Pages: A summary of the way in which the workload is used in each test scenario is given in the Performance Headlines section. This includes a more detailed description of the workload, hardware and software specifications. Glossary Page: 54 A short glossary of the terms used in the tables throughout this document. Page VII

8 CONTENTS 1 Overview Performance Headlines Local Queue Manager Test Scenario Nonpersistent Messages Local Queue Manager Persistent Messages Local Queue Manager Client Channels Test Scenario Nonpersistent Messages Client Channels Persistent Messages Client Channels Client Channels Distributed Queuing Test Scenario Nonpersistent Messages Server Channels Persistent Messages Server Channels Effect of Antivirus Software Large Messages MQI Response Times: 50bytes to 100MB Local Queue Manager bytes to 32KB KB to 2MB MB to 100MB K Messages Local Queue Manager Client Channel Distributed Queuing K Messages Local Queue Manager Client Channel Distributed Queuing MB Messages Local Queue Manager Client Channel Distributed Queuing Application Bindings Local Queue Manager Nonpersistent Messages Persistent Messages Client Channels Nonpersistent Messages Persistent Messages Distributed Queuing Nonpersistent Messages Persistent Messages Short & Long Sessions Performance and Capacity Limits... Error! Bookmark not defined. 7 Tuning Recommendations Tuning the Queue Manager Queue Disk, Log Disk, and Message Persistence Log Buffer Size, Log File Size, and Number of Log Extents Channels: Process or Thread, Standard or Fastpath? Applications: Design and Configuration Standard or Fastpath? Parallelism, Batching, and Triggering Measurement Environment Workload description MQI response time tool Test scenario workload Hardware Software Glossary...55

9 TABLES Table 1 Performance headline, nonpersistent messages, local queue manager...3 Table 2 Performance headline, persistent messages, local queue manager...4 Table 3 Performance headline, nonpersistent messages, client channels...6 Table 4 Performance headline, persistent messages, client channels...7 Table 5 1 round trip per driving application per second, client channels...9 Table 6 Free memory, client channels...11 Table 7 Performance headline, nonpersistent messages, server channels...13 Table 8 Performance headline, persistent messages, server channels...14 Table 9 1 round trip per driving application per second, client channels...16 Table 9 20K nonpersistent messages, local queue manager...22 Table 10 20K persistent messages, local queue manager...23 Table 11 20K nonpersistent messages, client channels...24 Table 12 20K persistent messages, client channels...25 Table 13 20K nonpersistent messages, client channels...26 Table 14 20K persistent messages, client channels...27 Table K nonpersistent messages, local queue manager...28 Table K persistent messages, local queue manager...29 Table K nonpersistent messages, client channels...30 Table K persistent messages, client channels...31 Table K nonpersistent messages, distributed queuing...32 Table K persistent messages, distributed queuing...33 Table 21 2M nonpersistent messages, local queue manager...34 Table 22 2M persistent messages, local queue manager...35 Table 23 2M nonpersistent messages, client channels...36 Table 24 2M persistent messages, client channels...37 Table 25 2M nonpersistent messages, distributed queuing...38 Table 26 2M persistent messages, distributed queuing...39 Table 27 Application binding, nonpersistent messages, local queue manager...40 Table 28 Application binding, persistent messages, local queue manager...41 Table 29 Application binding, nonpersistent messages, client channels...42 Table 30 Application binding, persistent messages, client channels...43 Table 31 Application binding, nonpersistent messages, distributed queuing...44 Table 32 Application binding, persistent messages, distributed queuing...45 Table 34 Short sessions, client channels...47 Page IX

10 FIGURES Figure 1 Connections into a local queue manager...2 Figure 2 Performance headline, nonpersistent messages, local queue manager...3 Figure 3 Performance headline, persistent messages, local queue manager...4 Figure 4 MQI-client channels into a remote queue manager...5 Figure 5 Performance headline, nonpersistent messages, client channels...6 Figure 6 Performance headline, persistent messages, client channels...7 Figure 7 1 round trip per driving application per second, client channels, nonpersistent messages...8 Figure 8 1 round trip per driving application per second, client channels, persistent messages...8 Figure 9 Free memory, client channels, nonpersistent messages...10 Figure 10 Free memory, client channels, persistent messages...10 Figure 11 Server channels between two queue managers...12 Figure 12 Performance headline, nonpersistent messages, server channels...13 Figure 13 Performance headline, persistent messages, server channels...14 Figure 14 1 round trip per driving application per second, server channel, nonpersistent messages..15 Figure 15 1 round trip per driving application per second, server channel, persistent messages...15 Figure 16 Effect of antivirus software, persistent messages, server channels...17 Figure 15 The effect of nonpersistent message size on MQI response time (50byte - 32K)...18 Figure 16 The effect of persistent message size on MQI response time (50byte - 32K)...19 Figure 17 The effect of nonpersistent message size on MQI response time (32K 2MB)...20 Figure 18 The effect of persistent message size on MQI response time (32K 2MB)...20 Figure 19 The effect of nonpersistent message size on MQI response time (2MB 100MB)...21 Figure 20 The effect of persistent message size on MQI response time (2MB 100MB)...21 Figure 21 20K nonpersistent messages, local queue manager...22 Figure 22 20K persistent messages, local queue manager...23 Figure 23 20K nonpersistent messages, client channels...24 Figure 24 20K persistent messages, client channels...25 Figure 25 20K nonpersistent messages, distributed queuing...26 Figure 26 20K persistent messages, distributed queuing...27 Figure K nonpersistent messages, local queue manager...28 Figure K persistent messages, local queue manager...29 Figure K nonpersistent messages, client channels...30 Figure K persistent messages, client channels...31 Figure K nonpersistent messages, distributed queuing...32 Figure K persistent messages, distributed queuing...33 Figure 33 2M nonpersistent messages, local queue manager...34 Figure 34 2M persistent messages, local queue manager...35 Figure 35 2M nonpersistent messages, client channels...36 Figure 36 2M persistent messages, client channels...37 Figure 37 2M nonpersistent messages, distributed queuing...38 Figure 38 2M persistent messages, distributed queuing...39 Figure 39 Application binding, nonpersistent messages, local queue manager...40 Figure 40 Application binding, persistent messages, local queue manager...41 Figure 41 Application binding, nonpersistent messages, client channels...42 Figure 42 Application binding, persistent messages, client channels...43 Figure 43 Application binding, nonpersistent messages, distributed queuing...44 Figure 44 Application binding, persistent messages, distributed queuing...45 Figure 46 Short sessions, client channels...46 Page X

11 1 Overview WebSphere MQ V6.0 on Windows 2003 has very similar performance characteristics to the V5.3 product. There are several graphs contained in this report that show specific tests have degraded by up to 5% but the majority of comparable measurements show similar or enhanced throughput. Page 1

12 2 Performance Headlines The measurements for the local queue manager scenario are for processing messages with no thinktime. For the client channel scenario and distributed queuing scenario, there are also measurements for rated messaging. No think-time is when the driving applications do not wait after getting a reply message before submitting subsequent request messages this is also referred to as tight-loop. The rated messaging tests used one round trip per driving application per second. In the client channel test scenarios, each driving application using a dedicated MQI-client channel, in the distributed queuing test scenarios, one or more applications submit messages over a fixed number of server channels. All tests are automatically stopped after the response time exceeds 1 second. 2.1 Local Queue Manager Test Scenario Input queue Driving applications Server application Reply queue Local queue manager Figure 1 Connections into a local queue manager 1) The driving application puts a message to the common input queue on the local queue manager, and holds on to the message identifier returned in the message descriptor. The driving application then waits indefinitely for a reply to arrive on the common reply queue. 2) The server application gets messages from the common input queue and places a reply to the common reply queue. The queue manager copies over the message identifier from the request message to the correlation identifier of the reply message. 3) The driving application gets a reply from the common reply queue using the message identifier held from when the request message was put to the common input queue, as the correlation identifier in the message descriptor. Nonpersistent and persistent messages were used in the local queue manager tests, with a message size of 2K. The effect of message throughput with larger messages sizes is investigated in the Large Messages section. Page 2

13 2.1.1 Nonpersistent Messages Local Queue Manager Figure 2 and Figure 3 shows the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the local queue manager scenario (see Figure 1 on the previous page), and WebSphere MQ V6.0 compared to Version 5.3. Figure 2 Performance headline, nonpersistent messages, local queue manager Note: Messaging in these tests is with no think-time. Figure 2 and Table 1 shows that the peak throughput of nonpersistent messages has increased by 6.8% (17,362 RT/s 18,539 RT/s) comparing Version 5.3 to Version 6.0. Test name: local_np WebSphere MQ V5.3 WebSphere MQ V6.0 4 (20) 4 (20) Round Trips/sec 17,362 (15,217) 18,539 (14,746) Response time (s) (0.0011) (0.0017) 76% (67%) 73% (61%) Table 1 Performance headline, nonpersistent messages, local queue manager Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 3

14 2.1.2 Persistent Messages Local Queue Manager Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 3 Performance headline, persistent messages, local queue manager Note: Messaging in these tests is with no think-time. Figure 3 and Table 2 show that the peak throughput of persistent messages has increased by 6.3% (3,188 RT/s 3,389 RT/s) comparing Version 5.3 to Version 6.0. Test name: local_pm Round Trips/sec Response time (s) WebSphere MQ V5.3 (108) 112 (120) (3,187) 3,188 (3,179) (0.0390) (0.0441) (39%) 39% (40%) WebSphere MQ V (112) (120) 3,389 (3,385) (3,388) (0.0372) (0.0429) 44% (44%) (44%) Table 2 Performance headline, persistent messages, local queue manager Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 4

15 2.2 Client Channels Test Scenario Driving application Server application 5 1 Input queue Client channel Driving machine Reply queue Server machine Remote queue manager Figure 4 MQI-client channels into a remote queue manager 1, 2) The driving application puts a request message (over a client channel), to the common input queue, and holds on to the message identifier returned in the message descriptor. The driving application then waits indefinitely for a reply to arrive on the common reply queue. 3) The server application gets messages from the common input queue and places a reply to the common reply queue. The queue manager copies over the message identifier from the request message to the correlation identifier of the reply message. 4, 5) The driving application gets the reply message (over the client channel), from the common reply queue. The driving application uses the message identifier held from when the request message was put to the common input queue, as the correlation identifier in the message descriptor. Nonpersistent and persistent messages were used in the client channel tests, with a message size of 2K. The effect of message throughput with larger messages sizes is investigated in the Large Messages section. Page 5

16 2.2.1 Nonpersistent Messages Client Channels Figure 5 and Figure 6 shows the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the client channel scenario (see Figure 4 on the previous page), and WebSphere MQ V6.0 compared to Version 5.3. Figure 5 Performance headline, nonpersistent messages, client channels Note: Messaging in these tests is with no think-time Figure 5 and Table 3 show that the peak throughput of nonpersistent messages has decreased by 2.7% (10,718 RT/s 9,904 RT/s) comparing Version 5.3 to Version 6.0. Test name: clnp WebSphere MQ V5.3 WebSphere MQ V6.0 (16) 17 (20) 16 (17) (20) Round Trips/sec (9,900) 10,718 (10,130) 9,904 (9,846) (9,838) Response time (s) (0.0014) (0.0024) (0.0022) (0.0026) (70%) 77% (84%) 75% (78%) (79%) Table 3 Performance headline, nonpersistent messages, client channels Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 6

17 2.2.2 Persistent Messages Client Channels Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 6 Performance headline, persistent messages, client channels Note: Messaging in these tests is with no think-time. Figure 6 and Table 4 show that the peak throughput of persistent messages has slightly improved (2,965 RT/s 3,061 RT/s) comparing Version 5.3 to Version 6.0. Test name: clpm Round Trips/sec Response time (s) WebSphere MQ V (120) 2,965 (2,954) (0.0432) (57%) 56% WebSphere MQ V (120) 3,061 (3,034) (0.0461) 59% (59%) Table 4 Performance headline, persistent messages, client channels Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 7

18 2.2.3 Client Channels For the following client channel measurements, the messaging rate used is 1 round trip per second per MQI-client channel, i.e. a request message outbound over the client channel and a reply message inbound over the channel per second. Figure 7 1 round trip per driving application per second, client channels, nonpersistent messages Note: Messaging in these tests is 1 round trip per driving application per second. Figure 8 1 round trip per driving application per second, client channels, persistent messages Figure 7, Figure 8 and Table 5 (next page) shows how WebSphere MQ V6.0 has a 3.3% degradation in the number of highly active MQI-client connections processing non-persistent messages into a single queue manager and a 10% degradation for persistent messages. Page 8

19 Test name: clnp_r3600 (WebSphereMQ v5.3) clpm_r3600 (WebSphereMQ v5.3) 4,950 (5,000) 150 (150) Rate/app/hr 3,600 3,600 Round Trips/sec 4,726 (4,886) 135 (150) Response time (s) (0.3679) (0.0458) Table 5 1 round trip per driving application per second, client channels 62% (47%) 17% (18%) Note: The large bold numbers in the table above show the WebSphere MQ V6.0 peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison with Version 5.3. Page 9

20 Figure 9 and Figure 10 shows the memory requirement of an MQI-client connection using the runmqlsr listener in WebSphere MQ V6.0. Figure 9 Free memory, client channels, nonpersistent messages Note: Messaging in these tests is 1 round trip per driving application per second Clients using non persistent messages consume 166K of Free memory in V6. Figure 10 Free memory, client channels, persistent messages Clients using Persistent messages consume 150K of Free memory in V6. This does effect the size of the Swap file but is not directly proportional to the amount of main memory needed because the page manger will discard unreferenced pages. Page 10

21 Test name: Free (MB) 100 / 4,950 3,191 / 2,626 (100 / 5,000) (3,188 / 2,636) clnp_r3600 (WebSphereMQ v5.3) clpm_r3600 (WebSphereMQ v5.3) 50 / 150 (50 / 150) 3,150 / 3,135 (3,156 / 3,144) Table 6 Free memory, client channels Note: The large bold numbers in the table above show the WebSphere MQ V6.0 peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison with Version 5.3. Note: The free memory shown in Table 6 represents the available real memory, not swap memory. The amount of Free used per channel is 110K bytes in MQ V53 and 116K bytes in MQ V6. For further calculations on the swap reservation and shared memory utilisation, refer to Error! Reference source not found.. This will be covered in the next version of this document. Page 11

22 2.3 Distributed Queuing Test Scenario Transmission queue per channel Input queue 1 2 Server channel Driving machine Local queue manager Transmission queue per channel Reply queue Server machine Remote queue manager Figure 11 Server channels between two queue managers 1) The driving application puts a message to a local definition of a remote queue located on the server machine, and holds on to the message identifier returned in the message descriptor. The driving application then waits indefinitely for a reply to arrive on a local queue. 2) The message channel agent takes messages off the channel and places them on the common input queue on the server machine. 3) The server application gets messages from the common input queue, and places a reply to the queue name extracted from the messages descriptor (the name of a local definition of a remote queue located on the driving machine). The queue manager copies over the message identifier from the request message to the correlation identifier of the reply message. 4) The message channel agent takes messages off the transmission queue and sends them over the channel to the driving machine. 5) The driving application gets a reply from a local queue. The driving application uses the message identifier held from when the request message was put to the local definition of the remote queue, as the correlation identifier in the message descriptor Nonpersistent and persistent messages were used in the distributed queuing tests, with a message size of 2K. The effect of message throughput with larger messages sizes is investigated in the Large Messages section. Page 12

23 2.3.1 Nonpersistent Messages Server Channels Figure 12 and Figure 13 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the distributed queuing scenario (see Figure 11 on the previous page), and WebSphere MQ V6.0 compared to Version 5.3. Figure 12 Performance headline, nonpersistent messages, server channels Note: Messaging in these tests is with no think-time. Figure 12 and Table 7 show that the peak throughput of nonpersistent messages is similar (12,804 RT/s 12,706 RT/s) comparing Version 5.3 to Version 6.0. Test name: dqnp WebSphere MQ V5.3 WebSphere MQ V (20) 19 (20) Round Trips/sec 12,804 (12,512) 12,706 (12,607) Response time (s) (0.0018) (0.0017) 78% (74%) 77% (74%) Table 7 Performance headline, nonpersistent messages, server channels Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 13

24 2.3.2 Persistent Messages Server Channels Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 13 Performance headline, persistent messages, server channels Note: Messaging in these tests is with no think-time Figure 13 and Table 8 show that the peak throughput of persistent messages has decreased by 1.5% (2,130 RT/s 2,098 RT/s) comparing Version 5.3 to Version 6.0. Test name: dqpm WebSphere MQ V5.3 WebSphere MQ V6.0 (290) (300) Round Trips/sec (2,113) 2,130 2,098 (2,093) Response time (s) (0.1363) (0.1692) (18%) 18% 19% (19%) Table 8 Performance headline, persistent messages, server channels Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 14

25 2.3.3 Server Channels For the following distributed queuing measurements, the messaging rate used is 1 round trip per driving application per second, i.e. a request message outbound over the sender channel, and a reply message inbound over the receiver channel per second. Note that there are a fixed number of 4 server channel pairs for the nonpersistent messaging tests, and 2 pairs for the persistent message tests. Figure 14 1 round trip per driving application per second, server channel, nonpersistent messages Note: Messaging in these tests is 1 round trip per driving application per second. Figure 15 1 round trip per driving application per second, server channel, persistent messages Figure 14, Figure 15 and Table 9 (next page) shows how WebSphere MQ V6.0 is similar in performance as version 5.3. Page 15

26 Test name: dqnp_r3600 (WebSphereMQ v5.3) dqpm_r3600 (WebSphereMQ v5.3) 5,450 (5,150) 2,050 (2,250) Rate/app/hr 3,600 3,600 Round Trips/sec 5,000 (5,000) 2,048 (2,120) Response time (s) (0.0064) (1.0743) Table 9 1 round trip per driving application per second, client channels 28% (31%) 17% (17%) Note: The large bold numbers in the table above show the WebSphere MQ V6.0 peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison with Version 5.3. Page 16

27 2.3.4 Effect of Antivirus Software on Server Channels Figure 16 shows the effect of antivirus software on persistent message throughput achieved using an increasing number of driving applications in the distributed queuing scenario. Figure 16 Effect of antivirus software, persistent messages, server channels Note: Messaging in these tests is with no think-time The entire suite of Benchmarks contained in this report was executed with both Anti-Virus Software enabled and disabled. Disabled means the Registry setting were modified like [HKEY_LOCAL_MACHINE\SOFTWARE\Intel\LANDesk\VirusProtect6\CurrentVersion\Storag es\filesystem\realtimescan] "OnOff"=dword: This is the only documented measurement while AV software was active. The effect on the other measurements averaged less than 2% but this measurement showed reduction in the peak messaging throughput rate of 40%. The DQ measurement causes the message to be queued twice in both the Requester and Responder Queue Manager and when more than 52 applications are running there is a high likelihood of messages being spilled to the file system rather than being held in the per Queue cache. Apparently, the AV software examines the data being passed to the file system. As the Queue depths get larger, the ability of the Queue Manager to process messages diminishes because of the additional processing load on the path to the file system. Page 17

28 3 Large Messages 3.1 MQI Response Times: 50bytes to 100MB Local Queue Manager Queue manager log configuration: LogPrimaryFiles=3, LogFilePages= bytes to 32KB Figure 17 show that the response time for MQPUT/GET pairs is similar for all nonpersistent message sizes between 50bytes and 32KB. Figure 17 The effect of nonpersistent message size on MQI response time (50byte - 32K) Page 18

29 Figure 18 show that the response for MQPUT/GET pairs is similar for all persistent message sizes between 50bytes and 32KB. Figure 18 The effect of persistent message size on MQI response time (50byte - 32K) Page 19

30 KB to 2MB Figure 19 show that the response time for MQPUT/GET pairs has improved for nonpersistent message sizes between 32KB and 2MB. Figure 19 The effect of nonpersistent message size on MQI response time (32K 2MB) Figure 20 show that the response for MQPUT/GET pairs is similar for all persistent message sizes between 32KB and 2MB. Figure 20 The effect of persistent message size on MQI response time (32K 2MB) Page 20

31 MB to 100MB Figure 21 show that the response time for MQPUT/GET pairs is slightly slower for all nonpersistent message sizes between 2MB and 100Mb. Figure 21 The effect of nonpersistent message size on MQI response time (2MB 100MB) Figure 22 show that the response for MQPUT/GET pairs is slightly slower for all persistent message sizes between 2MB and 100MB. Figure 22 The effect of persistent message size on MQI response time (2MB 100MB) Page 21

32 3.2 20K Messages Local Queue Manager Figure 23 and Figure 24 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the local queue manager scenario Nonpersistent Messages Figure 23 20K nonpersistent messages, local queue manager Figure 23 and Table 10 show that the peak throughput of nonpersistent messages is similar (7,966 RT/s 7,939 RT/s) comparing Version 5.3 to Version 6.0. Test name: local_np_20k WebSphere MQ V5.3 WebSphere MQ V6.0 2 (7) (20) (2) 7 (20) Round Trips/sec 7,966 (7,002) (5,906) (5,832) 7,939 (6,890) Response time (s) (0.0012) (0.0047) (0.0004) (0.0032) Table 10 20K nonpersistent messages, local queue manager 30% (98%) (89% (29%) 97% (95%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3 Page 22

33 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 24 20K persistent messages, local queue manager Figure 24 and Table 11 show that the peak throughput of persistent messages has increased by 9.2% (363 RT/s 397 RT/s) comparing Version 5.3 to Version 6.0. Test name: local_pm_20k WebSphere MQ V5.3 4 (100) WebSphere MQ V6.0 4 (92) Round Trips/sec 363 (90) 397 (86) Response time (s) (1.3059) (1.2590) Table 11 20K persistent messages, local queue manager 5% (21%) 5% (21%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 23

34 3.2.2 Client Channel Figure 25 and Figure 26 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the client channel scenario Nonpersistent Messages Figure 25 20K nonpersistent messages, client channels Figure 25 and Table 12 show that the peak throughput of nonpersistent messages is similar (3,519 RT/s 3,527 RT/s) comparing Version 5.3 to Version 6.0. Test name: clnp_20k WebSphere MQ V5.3 WebSphere MQ V (18) (20) (17) 18 (20) Round Trips/sec 3,519 (3,516) (3,452) (3,522) 3,527 (3,510) Response time (s) (0.0062) (0.0065) (0.0060) (0.0065) Table 12 20K nonpersistent messages, client channels 69% (71%) (73%) (73%) 74% (76%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 24

35 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 26 20K persistent messages, client channels Figure 26 and Table 13 show that the peak throughput of persistent messages has increased by 7.1% (321 RT/s 344 RT/s) comparing Version 5.3 to Version 6.0. Test name: clpm_20k WebSphere MQ V5.3 4 (88) WebSphere MQ V6.0 4 (84) Round Trips/sec 321 (76) 344 (72) Response time (s) (1.3615) (1.3947) Table 13 20K persistent messages, client channels 7% (18%) 7% (17%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 25

36 3.2.3 Distributed Queuing Figure 27 and Figure 28 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the distributed queuing scenario Nonpersistent Messages Figure 27 20K nonpersistent messages, distributed queuing Figure 27 and Table 14 show that the peak throughput of nonpersistent messages has decreased by 1.7% (3,796 RT/s 3,732 RT/s) comparing Version 5.3 to Version 6.0. Test name: Round Response dqnp_20k Trips/sec time (s) WebSphere MQ V , % WebSphere MQ V , % Table 14 20K nonpersistent messages, client channels Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. Page 26

37 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 28 20K persistent messages, distributed queuing Figure 28 and Table 15 show that the peak throughput of nonpersistent messages has decreased by 1.1% (379 RT/s 375 RT/s) comparing Version 5.3 to Version 6.0. Test name: dqpm_20k WebSphere MQ V5.3 WebSphere MQ V6.0 (104) (120) Round Trips/sec (377) (374) Response time (s) (0.3409) (0.3686) Table 15 20K persistent messages, client channels (8%) 7% 8% (7%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. Page 27

38 K Messages Local Queue Manager Figure 29 and Figure 30 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the local queue manager scenario Nonpersistent Messages Figure K nonpersistent messages, local queue manager Figure 29 and Table 16 show that the peak throughput of nonpersistent messages has decreased by 5.3% (139 RT/s 132 RT/s) comparing Version 5.3 to Version 6.0. Test name: local_np_200k WebSphere MQ V5.3 WebSphere MQ V6.0 9 (19) (20) (9) 19 (20) Round Trips/sec 139 (138) (138) (127) 132 (131) Response time (s) (0.1661) (0.1737) (0.0829) (0.1834) Table K nonpersistent messages, local queue manager 24% (24%) (24%) (24%) 24% (24%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3 Page 28

39 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure K persistent messages, local queue manager Figure 30 and Table 17 show that the peak throughput of persistent messages has increased by 11.7% (46 RT/s 51 RT/s) comparing Version 5.3 to Version 6.0. Test name: local_pm_200k WebSphere MQ V5.3 4 (56) WebSphere MQ V6.0 4 (56) Round Trips/sec 46 (42) 51 (42) Response time (s) (1.5467) (1.5839) Table K persistent messages, local queue manager 11% (13%) 10% (13%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 29

40 3.3.2 Client Channel Figure 31 and Figure 32 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the client channel scenario Nonpersistent Messages Figure K nonpersistent messages, client channels Figure 31 and Table 18 show that the peak throughput of nonpersistent messages is similar (286 RT/s 288 RT/s) comparing Version 5.3 to Version 6.0. Test name: clnp_200k WebSphere MQ V5.3 (19) 20 Round Trips/sec (285) 286 Response time (s) (0.0782) Table K nonpersistent messages, client channels WebSphere MQ V (20) (40%) 39% 38% 39% Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. Page 30

41 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure K persistent messages, client channels Figure 32 and Table 19 show that the peak throughput of persistent messages has increased by 6.4% (43 RT/s 46 RT/s) comparing Version 5.3 to Version 6.0. Test name: clpm_200k WebSphere MQ V5.3 4 (56) WebSphere MQ V6.0 4 (56) Round Trips/sec 43 (38) 46 (40) Response time (s) (1.7727) (1.6112) Table K persistent messages, client channels 13% (14%) 12% (15%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 31

42 3.3.3 Distributed Queuing Figure 33 and Figure 34 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the distributed queuing scenario Nonpersistent Messages Figure K nonpersistent messages, distributed queuing Figure 33 and Table 20 show that the peak throughput of nonpersistent messages has decreased by 1.4% (326 RT/s 322 RT/s) comparing Version 5.3 to Version 6.0. Test name: dqnp_200k WebSphere MQ V5.3 WebSphere MQ V (17) (20) (12) 17 (20) Round Trips/sec 326 (324) (326) (313) 322 (319) Response time (s) (0.0614) (0.0751) (0.0446) (0.0731) Table K nonpersistent messages, distributed queuing 44% (44%) (44%) (41%) 44% (44%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 32

43 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure K persistent messages, distributed queuing Figure 34 and Table 21 show that the peak throughput of nonpersistent messages has increased by 17.4% (33 RT/s 39 RT/s) comparing Version 5.3 to Version 6.0. Test name: dqpm_200k WebSphere MQ V (48) (32) Round Trips/sec 33 (31) (36) Response time (s) (1.8089) (1.0624) WebSphere MQ V Table K persistent messages, distributed queuing 4% (4%) 5% (5%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 33

44 3.4 2MB Messages Local Queue Manager Figure 35 and Figure 36 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the local queue manager scenario Nonpersistent Messages Figure 35 2M nonpersistent messages, local queue manager Figure 35 and Table 22 show that the peak throughput of nonpersistent messages is similar when comparing Version 5.3 to Version 6.0. Test name: local_np_2m WebSphere MQ V5.3 WebSphere MQ V6.0 6 (9) (11) (6) 9 (11) Round Trips/sec 7 (7) (7) (6) 7 (7) Response time (s) (1.5277) (1.8606) (1.1784) (1.8827) Table 22 2M nonpersistent messages, local queue manager 7% (7%) (7%) (7%) 8% (8%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3 Page 34

45 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 36 2M persistent messages, local queue manager Figure 36 and Table 23 show that the peak throughput of persistent messages is similar when comparing Version 5.3 to Version 6.0. Test name: local_pm_2m WebSphere MQ V5.3 4 (24) WebSphere MQ V6.0 4 (24) Round Trips/sec 3 (3) 3 (3) Response time (s) (9.6820) (9.6338) Table 23 2M persistent messages, local queue manager 4% (4%) 4% (4%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 35

46 3.4.2 Client Channel Figure 37 and Figure 38 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the client channel scenario Nonpersistent Messages Figure 37 2M nonpersistent messages, client channels Figure 37 and Table 24 show that the peak throughput of nonpersistent messages is similar when comparing Version 5.3 to Version 6.0. Test name: clnp_2m WebSphere MQ V5.3 WebSphere MQ V6.0 7 (9) (10) (7) 9 (10) Round Trips/sec 7 (7) (7) (6) 7 (7) Response time (s) (1.5386) (1.7083) (1.3420) (1.6990) Table 24 2M nonpersistent messages, client channels 11% (11%) (11%) (9%) 9% (10%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 36

47 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 38 2M persistent messages, client channels Figure 38 and Table 25 show that the peak throughput of persistent messages is similar when comparing Version 5.3 to Version 6.0. Test name: clpm_2m WebSphere MQ V5.3 (4) 24 Round Trips/sec (3) 3 Response time (s) (1.4494) (3) (9.4293) Table 25 2M persistent messages, client channels WebSphere MQ V6.0 4 (24) (5%) 6% 5% (6%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 37

48 3.4.3 Distributed Queuing Figure 39 and Figure 40 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the distributed queuing scenario Nonpersistent Messages Figure 39 2M nonpersistent messages, distributed queuing Figure 39 and Table 26 show that the peak throughput of nonpersistent messages is similar when comparing Version 5.3 to Version 6.0. Test name: dqnp_2m WebSphere MQ V5.3 WebSphere MQ V6.0 (10) (12) Round Trips/sec (7) 7 7 (7) Response time (s) (1.6068) (1.9338) Table 26 2M nonpersistent messages, distributed queuing (11%) 11% 11% (11%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. Page 38

49 Persistent Messages Queue manager log configuration for Version 5.3: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Queue manager log configuration for Version 6.0: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=4096 Figure 40 2M persistent messages, distributed queuing Figure 40 and Table 27 show that the peak throughput of nonpersistent messages is similar when comparing Version 5.3 to Version 6.0. Test name: dqpm_2m WebSphere MQ V5.3 (16) 24 Round Trips/sec (2) 2 Response time (s) (9.9738) (3) (9.8052) Table 27 2M persistent messages, distributed queuing WebSphere MQ V (24) (4%) 4% 5% (6%) Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 39

50 4 Application Bindings This report analyzes the rate that messages can be exchanged between a Requester (Driver) application and a Responder (Server) application. The other chapters use a Trusted Requester and a Shared Responder but this chapter looks at the effect of various combinations of application bindings for Requester and Responder programs. Requester Responder Normal Trusted Shared Isolated Isolated Isolated Trusted Trusted Trusted Non Trusted Shared Shared 4.1 Local Queue Manager Figure 41 and Figure 42 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the local queue manager scenario Nonpersistent Messages Figure 41 Application binding, nonpersistent messages, local queue manager Figure 41 and Table 28 show that the peak throughput of nonpersistent messages when comparing Normal, Isolated, Trusted and NonTrusted bindings Test Normal Isolated Trusted 4 (20) 5 (20) 6 (20) Round Trips/sec 18,539 (14,746) 6,598 (6,251) 31,649 (26,850) Response time (s) (0.0017) (0.0035) (0.0009) 73% (61%) 98% (98%) 90% (98%) NonTrusted 5 14, % (20) (11,785) (0.0021) (81%) Table 28 Application binding, nonpersistent messages, local queue manager Page 40

51 4.1.2 Persistent Messages Queue manager log configuration: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Figure 42 Application binding, persistent messages, local queue manager Figure 42 and Table 29 show that the peak throughput of persistent messages when comparing Normal, Isolated and Trusted bindings. Test Normal Isolated 108 (120) 84 (120) Round Trips/sec 3,389 (3,388) 2,518 (2,453) Response time (s) (0.0429) (0.0578) 44% (44%) 85% (85%) Trusted 120 3, % Table 29 Application binding, persistent messages, local queue manager Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 41

52 4.2 Client Channels Figure 43 and Figure 44 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the client channel scenario Nonpersistent Messages Figure 43 Application binding, nonpersistent messages, client channels Figure 43 and Table 30 show that the peak throughput of nonpersistent messages when comparing Normal, Isolated and Trusted bindings. Test Normal Isolated 16 (20) 17 (20) Round Trips/sec 9,904 (9,838) 9,807 (9,679) Response time (s) (0.0026) (0.0026) 75% (79%) 78% (79%) Trusted 20 11, % Table 30 Application binding, nonpersistent messages, client channels Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 42

53 4.2.2 Persistent Messages Queue manager log configuration: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Figure 44 Application binding, persistent messages, client channels Figure 44 and Table 31 show that the peak throughput of persistent messages when comparing Normal, Isolated and Trusted bindings. Test Normal Isolated 108 (120) 80 (120) Round Trips/sec 3,061 (3,034) 3,006 (2,941) Response time (s) (0.0461) (0.0484) 59% (59%) 60% (60%) Trusted 108 3, % (120) (3,208) (0.0437) (52%) Table 31 Application binding, persistent messages, client channels Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. Page 43

54 4.3 Distributed Queuing Figure 44 and Figure 45 show the nonpersistent and persistent message throughput achieved using an increasing number of driving applications in the distributed queuing scenario Nonpersistent Messages Figure 45 Application binding, nonpersistent messages, distributed queuing Figure 45 and Table 32 show that the peak throughput of nonpersistent messages when comparing Normal, Isolated and Trusted bindings. Test Normal Isolated 19 (20) 11 (20) Round Trips/sec 12,706 (12,607) 6,289 (5,686) Response time (s) (0.0017) (0.0036) 77% (74%) 25% (26%) Trusted 20 15, % Table 32 Application binding, nonpersistent messages, distributed queuing Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 44

55 4.3.2 Persistent Messages Queue manager log configuration: LogPrimaryFiles=4, LogFilePages=16384, LogBufferPages=512 Figure 46 Application binding, persistent messages, distributed queuing Figure 46 and Table 33 show that the peak throughput of persistent messages when comparing Normal, Isolated and Trusted bindings. Test Round Trips/sec Response time (s) Normal 120 1, % Isolated 120 1, % Trusted 120 1, % Table 33 Application binding, persistent messages, distributed queuing Note: The large bold numbers in the table above show the peak number of round trips per second, and the number of driving applications used to achieve the peak throughput. The numbers in brackets are included in the table to provide meaningful comparison between WebSphere MQ V6.0 and Version 5.3. Page 45

56 5 Short & Long Sessions The previous chapters in this report only reported on steady state messaging that does not include any session setup and termination function. This chapter specifically bracket groups of five MQPUT/MQGET pairs with MQCON/MQDISC and MQOPEN/MQCLOSE calls so a comparison of this overhead can be seen. A short session is a term used to describe the behaviour of an MQI application as it processes a small number of messages using one or more queues and a queue manager. The measurements in this document use an MQI-client application and the following sequence: connects to the queue manager opens the common input queue, and common reply queue puts a request message to the common input queue 5x gets the reply message from the common reply queue wait one second closes both queues disconnects from the queue manager Why measure short sessions? For each new connecting application or disconnecting application, the queue manager and Operating System must start a new process or thread and set up the new connection. As the number of connecting and disconnecting applications increases, the Operating System and queue manager are subjected to a higher load. While these requests are being serviced the queue manager has less time available to process messages, so fewer driving applications can be reconnected to the queue manager per second before the response time exceeds one second. This effect is greater than that of reducing the total messaging throughput of the queue manager by connecting thousands of MQI applications to the queue manager (refer to Figure 47 for an illustration). Figure 47 Short sessions, client channels Page 46