MQSeries.net :: View topic

goffinf · Posted: Sun Jan 31, 2010 2:36 pm Post subject:

Environment:

Broker version 6.1.0.3
MQ v6
Platform AIX

I have been asked to look at a number flows that are experiencing some performance and scalability issues and the occasional unexplained crash !

There are a number of fairly apparent areas of concern but before I go stamping around changing this that and the other I want to establish the current set of flows and their resource use profile as a baseline and go from there making incremental changes and reprofiling to assess the impact of each change.

I am certainly no expert here so I wanted to know what people typically use to profile both the flows and the resources used (at least memory, cpu, io, threads).

Since this is a new subject to me I'm not even sure yet what tooling and other resources we have available either as part of Broker, MQ or platform tools.

Well, obviously in Broker we have flowstats and tracing, but I can't say that I'm certain which might prove most useful. We *may* have some Tivoli agents(not sure which yet) and in the past we have used some NMon scripts provided by IBM. I had a scoot through the suppport packs yesterday to see what their might be to help both 'drive' testing and harvest results. Obvious perfromance reports for my environment are going to be worth reading through and a few reports from people like Tim Dunn, but beyond that, I obviously need to be able to understand the specifics of my environment and my flows.

I will have other people from our test teams to assist me (they are not Broker experts), but I want to get a basic feel first for the things I should be looking at (flows, machine resources, queues, broker database, Java settings, etc..) and how best to make sense of the data gathered to propose some changes and then measure the effect. I know enough to understand that 'guessing' is ok as a starting point but no more than that.

In some cases I certainly will want to test the flows under load and for reasonable periods of time (there is some suggestion of memory leakage for example, which occurs gradually but has an increasing effect). That might mean that there is a lot of data gathered (and I need to understand the overhead of that processing too).

I would really welcome comments from anyone who does or has done this kind of work before in terms of their approach and the tooling support.

Just by way of a simple resume' of the flows under scrutiny :-

The flows are used for relatively high thruput (circa: 15 message/second) and make numerous calls out to external [web] services both within and outside the organsation. They don't [currently] make a great deal of use of MQ (although that may change). The logic contained is of moderate to high complexity. Some flows contain custom Java.

Many thanks, all comments welcome.

Regards

Fraser.