|
RSS Feed - WebSphere MQ Support
|
RSS Feed - Message Broker Support
|
  |
|
Execution groups/threads/Solaris 10 |
View previous topic :: View next topic |
Author |
Message
|
andrewjhenderson |
Posted: Mon Dec 08, 2008 6:50 am Post subject: Execution groups/threads/Solaris 10 |
|
|
Novice
Joined: 06 Sep 2007 Posts: 15
|
You have MQ v6, WMB 6, and the largest Solaris 10 box that an architect could spec ( 32 processors and a zillion GB).
You deploy a compute intensive message flow with a reasonable number of additional threads ( say 20 for fun ).
Your expectation ( based purely on your life to date, without reading all the Guides to Multi-threading on sun.com that you no doubt should have) is that you load this thing up with messages of a reasonable size and the cpu on which it runs will get very busy , so that when you run mpstat, one cpu will say :
CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
0 118 0 2610 404 210 128 5 22 12548 1 22765 90 10 0 0
( or something similar but basically plenty of usr activity and 0 idl).
Having achieved max throughput on a single processor, you could then add a second execution group to load a second processor.
In this case, in fact mpstat says that eveything is fine , thanks, no cpu is particularly busy but they are all doing something.
However you then run ps auxww and you get
PID %CPU %MEM SZ RSS TT S START TIME COMMAND
24957 23.9 1.81 45521181848 ? O Dec 06 65:11 DataFlowEngine
23.9 % of the cpu on your 32 processor machine is in use!
Fantastic, you bought this huge box, and you can use it 1/4 of it with your single message flow with extra threads. ( like a virus!!).
Well in some ways that is good because it means there is transparent scaling for multi-threading process across multi-processors simultaneously ( see I did read sun.com).
However if you have been working in an MB world for a while, and assuming that an execution group process maps to a single processor at any one time ( and naturally governs the amount of resource you can soak up ), it is something to be aware of.
And I cant see anything about it in Solaris performance reports for MB, or in Tim Dunn's stuff about performance.
Is it just me? Or did everything change and I didnt notice?
Or does it just apply to Solaris 10?
I will now select an appropriate emoticon :  |
|
Back to top |
|
 |
andrewjhenderson |
Posted: Mon Dec 08, 2008 7:07 am Post subject: Answering my own post |
|
|
Novice
Joined: 06 Sep 2007 Posts: 15
|
As usual I find the right reference after I've posted.
Here it is for anyone else
p62 on Broker 6.1 Solaris SPARC performace support pac.
ftp://ftp.software.ibm.com/software/integration/support/supportpacs/individual/ip6f.pdf
Quote :
The T2000 machine used in this report contains an 8 core processor and can support up to 32 simultaneous processing threads. This “CoolThreads” technology is unique to this platform and has a distinct impact on Message Broker scaling behaviour. From a Message
Broker point of view this means that the broker can only utilise only up to 1/32 of the machine with a single flow instance i.e, 3.1% CPU.
Basically it scales beyond a single processor on this hardware/software combination.
So there you go. |
|
Back to top |
|
 |
|
|
  |
|
Page 1 of 1 |
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
|
|
|