| Author |
Message
|
| elikatz |
 Posted: Tue Apr 27, 2010 10:08 am Post subject: MQ low latency |
 |
|
Voyager
Joined: 24 Feb 2009
Posts: 86
|
hi all,
is anyone familiar with this product or has experience (good or bad...) with it?
we are currently using MQ version 6.0.2.8 on windows cluster but lately we notice an increase in the volume and requirement for low latency.
does it worth consider using this product?
(we are a software company in the financial industry if it helps with something..)
thanks! |
|
| Back to top |
|
 |
| gbaddeley |
| Posted: Tue Apr 27, 2010 4:22 pm Post subject: |
|
|
Jedi Knight
Joined: 25 Mar 2003
Posts: 2538
Location: Melbourne, Australia
|
MQ Low Latency is a separate product to MQ. It is very fast, but lacks a lot of the features of mainstream MQ. There would probably be more benefit in analysing your usage of MQ and working on the resource and application bottlenecks. MQ can achieve milllisecond level latency. Do you require sub-millisecond latency?
_________________
Glenn |
|
| Back to top |
|
|
| elikatz |
| Posted: Thu Apr 29, 2010 6:15 pm Post subject: |
|
|
Voyager
Joined: 24 Feb 2009
Posts: 86
|
thanks for your answer - i've consulted our dev team who advice that there are 2 things relevant to us:
1. the milliseconds level latency
2. they want to avoid the need of getting acknoledge got each message (avoid the roundtrip) - would the low latency product solve it? does it work with UDP?
thanks again.
Eli |
|
| Back to top |
|
|
| fjb_saper |
| Posted: Thu Apr 29, 2010 8:44 pm Post subject: |
|
|
Grand High Poobah
Joined: 18 Nov 2003
Posts: 20756
Location: LI,NY
|
| elikatz wrote: |
thanks for your answer - i've consulted our dev team who advice that there are 2 things relevant to us:
1. the milliseconds level latency
2. they want to avoid the need of getting acknowledge got each message (avoid the roundtrip) - would the low latency product solve it? does it work with UDP?
thanks again.
Eli |
Be more specific and very clear about point 2. I'm not getting what you mean by round trip and acknowledge...
AFAIK it only works with TCP as UDP would make it too unreliable. Think guaranteed delivery... 
I think there is somewhere a complete misconception and a need for re-architecting your project... but we cannot say without you giving us more information.
_________________
MQ & Broker admin |
|
| Back to top |
|
|
| exerk |
| Posted: Fri Apr 30, 2010 12:21 am Post subject: |
|
|
Jedi Council
Joined: 02 Nov 2006
Posts: 6339
|
| elikatz wrote: |
thanks for your answer - i've consulted our dev team who advice that there are 2 things relevant to us:
1. the milliseconds level latency |
Investigate why they want such low latency - it's not unusual for someone in an ivory tower somewhere to specify millisecond network time which is completely negated by multi-second application processing time...
| elikatz wrote: |
| 2. they want to avoid the need of getting acknoledge got each message (avoid the roundtrip) - would the low latency product solve it? |
I'm with fjb_saper on this one; what round trip? If the developers don't want 'acknowledgements' then they should not be a) doing request/reply, or b) setting report options.
_________________
It's puzzling, I don't think I've ever seen anything quite like this before...and it's hard to soar like an eagle when you're surrounded by turkeys. |
|
| Back to top |
|
|
| bruce2359 |
| Posted: Fri Apr 30, 2010 4:53 am Post subject: |
|
|
Poobah
Joined: 05 Jan 2008
Posts: 9469
Location: US: west coast, almost. Otherwise, enroute.
|
What are the exact requirements? How many milliseconds for request-reply to complete?
What are your bandwidth capabilities?
How well provisioned are your server boxes?
If there is no requirement for reply messages, then use message-type datagram messages.
If your app design is to include COA and COD messages, search here for lots of discussion on their real vs. perceived value.
There are often tuning opportunities on the hardware, o/s and mq to maximize throughput.
_________________
I like deadlines. I like to wave as they pass by.
ב''ה
Lex Orandi, Lex Credendi, Lex Vivendi. As we Worship, So we Believe, So we Live. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Fri Apr 30, 2010 9:55 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
You want MQ 7, which allows you to do multiple MQPUTs without waiting for a reply on the results of that MQPUT. You assume they are all working. Every so often you can issue a call to see how your MQPUTs are doing so far.
Obviously this is a lower class of service, but if its good enough you can get some insanely good #s, assuming the rest of the design is corrrect.
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| elikatz |
| Posted: Fri Apr 30, 2010 11:26 am Post subject: |
|
|
Voyager
Joined: 24 Feb 2009
Posts: 86
|
all,
thanks for all the replies.
i talked to one of our dev directors and he told me that he would be able to give the answers to all the questions once they start to do more thorough tests.
in the meanwhile, as Peter suggested I'll try to push MQ 7 since MQ will be end of service next September - that's a good way to push MQ 7.
(Thanks Peter!  )
thanks all, next week i'll install MQ 7.0.1 and see how it goes. |
|
| Back to top |
|
|
| mvic |
| Posted: Fri Apr 30, 2010 2:30 pm Post subject: |
|
|
Jedi
Joined: 09 Mar 2004
Posts: 2080
|
| PeterPotkay wrote: |
| Obviously this is a lower class of service, but if its good enough you can get some insanely good #s, assuming the rest of the design is corrrect. |
And some very full queues |
|
| Back to top |
|
|
| JosephGramig |
| Posted: Mon May 03, 2010 4:57 am Post subject: |
|
|
Grand Master
Joined: 09 Feb 2006
Posts: 1244
Location: Gold Coast of Florida, USA
|
| Well, if performance or reliability was a goal, then I would not use Windows as an OS. |
|
| Back to top |
|
|
| elikatz |
| Posted: Mon May 03, 2010 8:20 pm Post subject: |
|
|
Voyager
Joined: 24 Feb 2009
Posts: 86
|
Do you suggest that Linux gives better performance?
Last time I thought of going with Linux I got into difficulties with finding easy HA solution for our production environment... but if it's much better then windows I'll reconsider... |
|
| Back to top |
|
|
| mvic |
| Posted: Tue May 04, 2010 3:12 am Post subject: |
|
|
Jedi
Joined: 09 Mar 2004
Posts: 2080
|
| elikatz wrote: |
| Do you suggest that Linux gives better performance? |
With the same CPU, memory and I/O, there should be no significant difference. The way you use the system (persistent messages, large messages, syncpoint etc.) make much more difference than the choice of OS ever will. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Tue May 04, 2010 4:01 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
I don't remember where I read it, maybe the Performance Report support packs, but given the exact same hardware and MQ version, Linux would perform better than Windows. The comparision is possible because you can install Linux or Windows on the exact same metal hardware.
I agree with mvic though that application design and implementation has a far greater impact on end results. But if you're looking to squeeze out every last bit of performance, I'd say start with Linux.
Of course a misconfigured Linux system will perform worse than a properly tuned Windows one. The skill of your respective Sys Admins and your comfort level with each O/S should probably come into the decision making process.
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| bruce2359 |
| Posted: Tue May 04, 2010 5:53 am Post subject: |
|
|
Poobah
Joined: 05 Jan 2008
Posts: 9469
Location: US: west coast, almost. Otherwise, enroute.
|
It's too simple to say that Linux performs better (more efficiently, faster, ...)than Windows. Yes, in a lab environment with identical hardware, where only the o/s and MQ are installed, a slight performance improvement can be measured and demonstrated.
However, real-life is different.
The questions to ponder in addressing:
Have you clearly demonstrated and documented (with your test case or production data) where the bottlenecks are? Without a goal, you are only guessing.
Where is time/money best spent in addressing performance?
Can similar performance improvement be gained in Windoze (or other o/s) tuning what you already have? Or by adding netowrk bandwidth, CPU, RAM or I/O, or something else?
Like most other things in life, o/s choice is a trade-off. Does your organization want to implement another o/s? Is it worthwhile to do so for a 5% performance improvement?
What if in 6 months or one year, performance degrades again? Will you recommend yet another o/s switch? I'd guess that this will deonstrate to management that you (and your performance tuning team) don't understand the real issues.
I would be remiss if I didn't mention that if your organization is looking for absolute and dramatic performance improvement, look to WMQ on z/OS. (I await the predictable replies ...)
_________________
I like deadlines. I like to wave as they pass by.
ב''ה
Lex Orandi, Lex Credendi, Lex Vivendi. As we Worship, So we Believe, So we Live. |
|
| Back to top |
|
|
|
|
