IBM MQ - better application performance
11 likes9,691 views
Presented in Feb 2015 at Interconnect This presentation is aimed at helping application developers understand how to best use MQ features for higher performance.
![[115]쿠팡 서비스 클라우드 마이그레이션 통해 배운것들](https://cdn.slidesharecdn.com/ss_thumbnails/115coupang-181011031522-thumbnail.jpg?width=560&fit=bounds)
![[오픈소스컨설팅] 쿠버네티스와 쿠버네티스 on 오픈스택 비교 및 구축 방법](https://cdn.slidesharecdn.com/ss_thumbnails/osck8svsk8sonopenstackkhoj-210310051504-thumbnail.jpg?width=560&fit=bounds)
![[AWS Dev Day] 실습워크샵 | Amazon EKS 핸즈온 워크샵](https://cdn.slidesharecdn.com/ss_thumbnails/awsdevdayseoul2019-amazonekshands-onworkshopv0-190930092336-thumbnail.jpg?width=560&fit=bounds)
Ad
More Related Content
What's hot (20)
Similar to IBM MQ - better application performance (20)
Ad
More from MarkTaylorIBM (10)
Ad
Recently uploaded (20)
![Pixologic ZBrush Crack Plus Activation Key [Latest 2025] New Version](https://cdn.slidesharecdn.com/ss_thumbnails/fashionevolution2-250322112409-f76abaa7-250428124909-b51264ff-250504160528-fc2bb1c5-thumbnail.jpg?width=560&fit=bounds)
IBM MQ - better application performance
- 1. © 2015 IBM Corporation 2279: IBM MQ Better Application Performance Mark Taylor [email protected] IBM Hursley
- 2. Please Note IBM’s statements regarding its plans, directions, and intent are subject to change or withdrawal without notice at IBM’s sole discretion. Information regarding potential future products is intended to outline our general product direction and it should not be relied on in making a purchasing decision. The information mentioned regarding potential future products is not a commitment, promise, or legal obligation to deliver any material, code or functionality. Information about potential future products may not be incorporated into any contract. The development, release, and timing of any future features or functionality described for our products remains at our sole discretion. Performance is based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput or performance that any user will experience will vary depending upon many factors, including considerations such as the amount of multiprogramming in the user’s job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve results similar to those stated here.
- 3. Agenda • Some performance numbers • Structure of the Queue Manager • Function Walkthroughs • Client Channels • Logging and Recovery
- 4. Agenda • Some performance numbers • Structure of the Queue Manager • Function Walkthroughs • Client Channels • Logging and Recovery
- 5. Performance Bottlenecks • Business Systems are complex ‒ Often no single bottleneck limiting performance ‒ Performance can mean different things to different people ‒ Throughput Scalability Low resource usage • Not only limited by physical resources ‒ Application design, such as parallelism can have major effect • Performance Reports (from SupportPac site) show range of scenarios RAM CPU NetworkI/O Bottlenecks
- 6. See how far it's come …
- 7. Distributed Performance: Persistent Messaging • An AIX system: Max performance increased from 16K to 19K
- 8. Distributed Performance: Non-persistent Messaging • A Windows system: Max performance increased from 90K to 121K
- 9. Distributed Performance in V8 • Improvements to distributed queue manager scaling ‒ Improve efficiency ‒ Better exploitation of wider SMP machines ‒ Looking at customer-provided application patterns not just benchmarks Always happy to have more customer examples • Multiplexed client performance ‒ Increase the performance of multiplexed client channels (SHRCONV > 0) ‒ Especially for SHRCONV=1 • Other areas that helped: ‒ Cache alignment for internal structures ‒ Extended 64-bit exploitation for locking primitives ‒ RFH2 handling, particularly for waiting-getter Fewer copies of data are needed ‒ Better compiler optimisations including feedback-directed optimisation ‒ Faster data conversion (especially for 1208) Many messages are in 1208 codepage Optimised handling when the queue manager needs to convert them
- 10. Agenda • What is distributed WebSphere MQ? • Structure of the Queue Manager • Function Walkthroughs • Client Channels • Logging and Recovery
- 11. Also: runmqchi runmqlsr amqzdmaa amqrmppa amqpcsea Execution Controller (amqzxma0) Critical Services (amqzmuc0) Restartable Services (amqzmur0) External Processes (amqzmgr0) Repository Manager (amqrrmfa) Object Authority Manager (amqzfuma) LQM Agent (amqzlaa0) Queue Manager Process Tree V6V7.x, V8 PubSub V6 Compat (amqfqpub) PubSub Streams (amqfcxba) PubSub Utils (amqzmuf0) Dur/Subs Mgr Topic Scavenger Multicast Brw/Mark Delayed Deliv Cache, Inter QMgr
- 12. Agent Queue Manager Shared Resources Checkpoint MQCONN MQCONN MQOPEN MQDISC MQPUT MQGET etc. Agent IPCC Execution Controller IPCC Log Manager MQI Stub Application IPCC MQI Stub Application IPCC MQI Stub Application IPCC Object Authority Manager Queue Manager: Process Model Shared, Isolated and Trusted Bindings
- 13. Agenda • What is distributed WebSphere MQ? • Structure of the Queue Manager • Function Walkthroughs • Client Channels • Logging and Recovery
- 14. MQI Stub Application IPCC Execution Controller IPCC IPCC DAP Kernel Agent IPCC Application (MQI Stub) Verify parameters and handles Construct a Connect message Call API crossing exit Send a message to the EC Application (MQI Stub) Receive the reply Construct an IPCC message Send the message to the agent Application (MQI Stub) Receive reply and call API crossing exit Return HCONN Execution Controller Choose an agent or start a new one Construct a reply IPCC message Return reply to application Agent Check App's permission to connect Allocate and assign agent resources Send IPCC reply back to application MQCONN
- 15. Performance Implications: Connection Binding • Fastpath binding removes inter-process communications ‒ Implies that the application is 'trusted' ‒ MQCONNX option MQCNO_FASTPATH_BINDING ‒ Application failure can corrupt queue manager • Primary benefit is for non-persistent message processing ‒ Use for MCAs, Broker - 30% CPU saving MQ Appl Agent Memory Log IPC Standard Binding MQ Appl Agent Memory Log FASTPATH Binding
- 16. Performance Implications: Connection Binding MQ Appl Agent Memory Log IPC Standard Binding MQ Appl Agent Memory Log FASTPATH Binding MCA Agent Memory Log IPC Client Binding MQ Appl Network
- 17. Application (MQI Stub) Verify hConn Call API crossing exit Verify parameter addressability Place parameters in an Open message Send an IPCC message to the agent MQI Stub Application IPCC DAP Kernel Agent IPCC Agent Application interface Verify open parameters Kernel Verify operation validity Resolve target – including cluster lookup Check permissions on the queue DAP Load the queue ready for gets and puts if required ƒThis is the part that can use system resources Kernel Generate handle to object for application Generate responses and event messages IPCC Send reply back to application Application (MQI Stub) Receive reply Call API crossing exit Return HOBJ MQOPEN of a queue
- 18. MQOPEN internal changes for V8 • The object catalog is the list of objects and a map to the real files ‒ Shows that queue 'X.Y' is stored in "/var/mqm/qmgrs/<qm>/queues/X!Y" ‒ dspmqfls command shows the map ‒ Updated whenever an object is created or deleted • In MQ V7 there were just two locks associated with the object catalog ‒ A significant implication of this scheme was that an MQOPEN could not overlap with the creation/deletion of a queue (or any other object) ‒ Ghost queues introduced in MQ V5.2 went some way to reducing the impact, particularly for TDQ’s, but very significant serialization implications remained. • MQ V8 uses finer grained locks ‒ The master mutex need not be owned while waiting for the object mutex. • The most significant impact of this serialization change is that dynamic queues can now be created and destroyed concurrently with MQOPEN activity 27
- 19. Performance Implications: Heavyweight MQI Calls • MQCONN is a “heavy” operation ‒ Don’t let your application do lots of them ‒ Wrappers and OO interfaces can sometimes hide what’s really happening ‒ Lots of MQCONNs can drop throughput from 1000s Msgs/Sec to 10s Msgs/Sec • MQOPEN is also ‘heavy’ compared to MQPUT/MQGET ‒ Depends on the type of queue and whether first use Loading pre-existing queue; creating dynamic queue ‒ It’s where we do the security check Try to cache queue handles if more than one message ‒ If you’re only putting one message consider using MQPUT1 Particularly client bindings • Try to avoid exclusive access to the Queue ‒ Makes it harder to scale the solution For example adding more instances of application ‒ Implies that reliance on message order is not required Partition the data to allow parallel processing?
- 20. 01 00 00 01 01 01 01 11 11 11 11 11 11 00 00 00 00 01 01 01 00 00 00 00 00 11 11 11 11 11 10 10 10 10 10 10 10 10 10 10 01 00 00 01 01 01 01 11 11 11 11 11 11 00 00 00 00 01 01 01 00 00 00 00 00 11 11 11 11 11 10 10 10 10 10 10 10 10 10 10 Space Map 01 00 00 01 01 01 01 11 11 11 11 11 11 00 00 00 00 01 01 01 00 00 00 00 00 11 11 11 11 11 10 10 10 10 10 10 10 10 10 10 Non- persistent Message Detail Cache Queue file Log Message Chain(s) Space Hdl File System Buffer Overflow Persistent Queue Buffer NP Queue Buffer Overflow Persistent In the Depths of a Queue
- 21. Tuning Queue Buffers • Increasing buffers can improve performance ‒ More information can be kept in memory, without flushing to disk ‒ Costs more memory per modified queue • But no documented external mechanism to do it ‒ Performance supportpacs indicate how to do it ‒ DefaultQBufferSize / DefaultPQBufferSize ‒ SupportPac MS0P (Cat2 – ie “as-is”) includes “QTune” program c:> java -jar qtune.jar -d c:mqmqmgrsQMAqueuesSYSTEM!DEFAULT!LOCAL!QUEUE File c:mqmqmgrsQMAqueuesSYSTEM!DEFAULT!LOCAL!QUEUEq Stored npBuff = 64 kB Stored pBuff = QMgr default Stored maxQSize = 2,097,151 MB
- 22. Performance Implications: Persistence • Log bandwidth is going to restrict throughput ‒ Put the log files on the fastest disks you have, separate from queue file ‒ Persistent messages are the main things requiring recovery after an outage ‒ Can significantly affect restart times • Why use persistence? ‒ False assumption that persistence is for "important" data and nonpersistent for when you don't care ‒ The real reason for persistent messages is to reduce application complexity ‒ With persistent, apps do not need logic to detect and deal with lost messages ‒ If your app (or operational procedures) can detect and deal with lost messages, then you do not need to use persistent messages • Understand which apps use persistence MQPUT Queue File Log
- 23. Kernel Verify operation validity (Resolve cluster queue destination) DAP Reserve space for the message data If (persistent message) Write log records for the update (Wait for log records to reach the disk if outside syncpoint) Write the message to the queue file Else (non-persistent) If (space available in queue buffer) Copy the message data into the buffer Else Write the message to the queue file without logging Maintain queue statistics such as queue depth Kernel Generate responses and events, wakeup getters/drive async consumers Serialised MQPUT Walkthrough Also check for "if waiting getter" Tip! Use persistent messages with syncpoint
- 24. Put to a waiting getter • MQPUT most efficient if there is getting application waiting ‒ Having multiple applications processing queue increases percentage ‒ May not appear ‘balanced’ – May keep one ‘hot’ • Only for out of syncpoint messages ‒ Both persistent and non-persistent ‒ If Persistent outside of syncpoint, think carefully about why using persistence Got message could be lost if crash before returning to the application! • No queuing required ‒ Removes a lot of processing of placing the message onto the queue • Significantly reduces CPU cost and improves throughput ‒ Lots of improvements on this area in newest releases (7.1, 7.5) MQPUT MQGET MQGET MQGET
- 25. Kernel Verify operation validity Check message expiry Wait for message if not available DAP Locate a message meeting the requested criteria including current browse cursor position priority message id, correlation id, segment or group conditions Copy data into the message buffer If (persistent) Write log record (Wait for log record to reach the disk if outside syncpoint) Move the browse cursor if required Maintain queue statistics such as queue depth Kernel Generate responses and events Serialised MQGET Walkthrough Tip! In request reply model, get by correlid more efficient than get by msgid
- 26. Kernel Verify operation validity DAP Write log record to end transaction Wait for this log record to reach the disk Lock all queues touched in this transaction For each queue Make any changes to messages in the transaction visible Unlock queue Kernel Generate responses and events Wakeup Getters/Drive Async Consumers MQCMIT Walkthrough Deal with 2PC protocol if in an XA transaction
- 27. Performance Implications: Syncpoint • Do you need it? ‒ Yes, when a set of work needs to either all be performed, or all not performed • Maximum Size of UOW can be limited ‒ QMGR MAXUMSGS parm ‒ Set to sensible value to avoid runaway applications • Make sure you keep the size of your UOWs small ‒ Don’t forget to end the UOW • Cheaper to process in syncpoint for persistent messages ‒ Up to a point, not huge UOWs ‒ Log not forced after every MQPUT/MQGET • Useful even when only a single message inside syncpoint ‒ And running multiple parallel applications
- 28. Good Application Design - Summary • Long-running connection • Opens queues up-front • Uses syncpoint for persistent operations • No message affinities so multiple instances can run in parallel
- 29. Publish/Subscribe Implementation from V7 • MQOPEN, MQPUT, MQGET very similar to point-to-point ‒ Includes cluster resolution ‒ Need to find closest admin topic node ‒ Internal subscribers may forward publication to another queue manager • Durable Subscriptions held on SYSTEM.DURABLE.SUBSCRIBER.QUEUE ‒ Multiple subscriptions consolidated into single message ‒ Why is there no non-durable subscriber queue? ‒ Retained publications also stored on a queue • Handling application abend ‒ V6 cleanup for non-durable subs was “automatic” for JMS, manual otherwise ‒ Automatic for V7+ • Managed destinations ‒ Agent creates queue in MQSUB - trace shows internal MQOPEN (kqiOpenModel) • Parallel match-space access via shared memory set ‒ Several applications can publish simultaneously on the same topic ‒ Lock held during subscribe/unsubscribe processing
- 30. Message Processing • Persistent pubs switch to non-persistent-ish for non-durable subscriptions ‒ Does not change the reliability level ‒ Messages are not logged, but they keep the “persistent” flag ‒ Improves performance • Properties stored as part of the message ‒ Logged for persistence, rcdmqimg etc ‒ Written to disk in either RFH2 or an “internal” format ‒ Converted to application-required format during MQGET • Selectors on queues can cause all messages to be browsed ‒ Queue lock may be held during selection
- 31. Agenda • What is distributed WebSphere MQ? • Structure of the Queue Manager • Function Walkthroughs • Client Channels • Logging and Recovery • Multiple Installation Support
- 32. Clients from V7 • Many changes to the client protocols in V7 ‒ All can improve performance • Multiplexing or Shared Conversations ‒ For multi-threaded applications ‒ Several connections use the same socket • Asynchronous Put ‒ For sending applications • Read-ahead ‒ For receiving applications • New threads inside application for full-duplex comms ‒ Sharecnv(0) – May be fast but no full-duplex so miss good functionality ‒ Sharecnv(1) – One socket per hconn, optimised in V8 - recommended value ‒ Sharecvn(10) – Shared socket for multiple conversations (default value)
- 34. Read-ahead of messages Client MQCONN MQOPEN MQGET MQGET MQGET Server Request for ‘n’ messages
- 35. Client Connection Performance. • Lots of internal changes at V8 ‒ Channel status table restructure ‒ Reduce cost of health checking (MQIBindType=FASTPATH) ‒ Keep a hash table of processes, each with a sub-list of threads. ‒ Reduce cost of creating/deleting TDQ’s ‒ Channel process pooling pre-start • Net effect of all of these changes is significant improvement to the queue manager’s ability to accept a large number of inbound connections in a relatively short time. • In one test the time taken to attach 50,000 clients, with each client owning a TDQ, reduced from over an hour at 7.0.1.6 to under 4 minutes at MQ V8. 59
- 36. Agenda • What is distributed WebSphere MQ? • Structure of the Queue Manager • Function Walkthroughs • Channels • Logging and Recovery • Multiple Installation Support
- 37. What's the point of logging? • A log record is written for each persistent update ‒ The log record describes the update • Optimisations to minimise serialisation points • Write-Ahead Logging ‒ The log is always more up-to-date than the actual data • Log is a sequential file ‒ Sequential I/O is much quicker than random ‒ Single point of writing rather than to individual object files • Log and actual data are reconciled during strmqm ‒ Progress information displayed • Point of consistency – Checkpoint ‒ Log control file: amqhlctl.lfh – in log directory ‒ Checkpoint amqalchk.fil – qmgr directory ‒ Backup queue managers from MQ V6 Myth! Triple write integrity does not mean we write all data 3 times!
- 38. Looking at logger performance • Can extract internal information from a service tool ‒ Lots of MQ performance PMRs turn out to be disk-related. So recording was added to the internal state amqldmpa –c H –m <qmgr> -d 8 –n <count> -s <interval> -f <file> ‒ The amqldmpa program can dump lots of other internal information too • Includes logger I/O activity ‒ WriteTimeMax shows maximum time (microseconds) to complete I/O ‒ WriteSizeMax shows largest (bytes) I/O ‒ Since qmgr started • Maintains averages ‒ WriteSizeShort is short-term (64-sample) weighted average of recent writes ‒ WriteSizeLong is longer-term (1024-sample) weighted average ‒ Similarly for WriteTimeShort/Long • From one PMR: ‒ WriteTimeMax = 59102377, WriteSizeMax=2097152 ‒ So it has taken nearly 60 seconds to write 2MB ‒ Implies they need to talk to disk support team!
- 39. Summary – understand the applications • Understand the application pattern ‒ Exclusivity, affinity • Understand the individual application ‒ Many times have heard "it's using persistence" when it's not. Or vice versa. • If you don't have access to application source, tools exist to help ‒ Application Activity Reports ‒ MQ Trace
- 40. Notices and Disclaimers Copyright © 2015 by International Business Machines Corporation (IBM). No part of this document may be reproduced or transmitted in any form without written permission from IBM. U.S. Government Users Restricted Rights - Use, duplication or disclosure restricted by GSA ADP Schedule Contract with IBM. Information in these presentations (including information relating to products that have not yet been announced by IBM) has been reviewed for accuracy as of the date of initial publication and could include unintentional technical or typographical errors. IBM shall have no responsibility to update this information. THIS DOCUMENT IS DISTRIBUTED "AS IS" WITHOUT ANY WARRANTY, EITHER EXPRESS OR IMPLIED. IN NO EVENT SHALL IBM BE LIABLE FOR ANY DAMAGE ARISING FROM THE USE OF THIS INFORMATION, INCLUDING BUT NOT LIMITED TO, LOSS OF DATA, BUSINESS INTERRUPTION, LOSS OF PROFIT OR LOSS OF OPPORTUNITY. IBM products and services are warranted according to the terms and conditions of the agreements under which they are provided. Any statements regarding IBM's future direction, intent or product plans are subject to change or withdrawal without notice. Performance data contained herein was generally obtained in a controlled, isolated environments. Customer examples are presented as illustrations of how those customers have used IBM products and the results they may have achieved. Actual performance, cost, savings or other results in other operating environments may vary. References in this document to IBM products, programs, or services does not imply that IBM intends to make such products, programs or services available in all countries in which IBM operates or does business. Workshops, sessions and associated materials may have been prepared by independent session speakers, and do not necessarily reflect the views of IBM. All materials and discussions are provided for informational purposes only, and are neither intended to, nor shall constitute legal or other guidance or advice to any individual participant or their specific situation. It is the customer’s responsibility to insure its own compliance with legal requirements and to obtain advice of competent legal counsel as to the identification and interpretation of any relevant laws and regulatory requirements that may affect the customer’s business and any actions the customer may need to take to comply with such laws. IBM does not provide legal advice or represent or warrant that its services or products will ensure that the customer is in compliance with any law.
- 41. Notices and Disclaimers (con’t) Information concerning non-IBM products was obtained from the suppliers of those products, their published announcements or other publicly available sources. IBM has not tested those products in connection with this publication and cannot confirm the accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on the capabilities of non-IBM products should be addressed to the suppliers of those products. IBM does not warrant the quality of any third-party products, or the ability of any such third-party products to interoperate with IBM’s products. IBM EXPRESSLY DISCLAIMS ALL WARRANTIES, EXPRESSED OR IMPLIED, INCLUDING BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. The provision of the information contained herein is not intended to, and does not, grant any right or license under any IBM patents, copyrights, trademarks or other intellectual property right. • IBM, the IBM logo, ibm.com, Bluemix, Blueworks Live, CICS, Clearcase, DOORS®, Enterprise Document Management System™, Global Business Services ®, Global Technology Services ®, Information on Demand, ILOG, Maximo®, MQIntegrator®, MQSeries®, Netcool®, OMEGAMON, OpenPower, PureAnalytics™, PureApplication®, pureCluster™, PureCoverage®, PureData®, PureExperience®, PureFlex®, pureQuery®, pureScale®, PureSystems®, QRadar®, Rational®, Rhapsody®, SoDA, SPSS, StoredIQ, Tivoli®, Trusteer®, urban{code}®, Watson, WebSphere®, Worklight®, X-Force® and System z® Z/OS, are trademarks of International Business Machines Corporation, registered in many jurisdictions worldwide. Other product and service names might be trademarks of IBM or other companies. A current list of IBM trademarks is available on the Web at "Copyright and trademark information" at: www.ibm.com/legal/copytrade.shtml.
- 42. Thank You Your Feedback is Important! Access the InterConnect 2015 Conference CONNECT Attendee Portal to complete your session surveys from your smartphone, laptop or conference kiosk.