MQSeries.net :: View topic - Broker on AIX and TCP delay

MQSeries.net

Tech Exchange

Education

Certifications

Library

Info Center

SupportPacs

FAQÂ Â

Usergroups

RSS Feed - WebSphere MQ Support

RSS Feed - Message Broker Support

MQSeries.net Forum Index » WebSphere Message Broker (ACE) Support » Broker on AIX and TCP delay

Broker on AIX and TCP delay

« View previous topic :: View next topic »

Author

Message

marcin.kasinski

Posted: Mon Mar 12, 2007 6:12 am Post subject: Broker on AIX and TCP delay

Sentinel

Joined: 21 Dec 2004
Posts: 850
Location: Poland / Warsaw

Message Broker v6 on AIX

Hi,

I have some question.

We had some problem with message flow.
In this flow we had HTTPRequest node.

During our tests we discovered very poor performance of this flow on AIX.
The same very simple flow on Win had 7-9 times better results.

We raised PMR.

IBM asked us to use command below to switch off AIX system TCP delay.

Code:

mqsichangeproperties <brokername> -b httplistener-o HTTPConnector -n tcpNoDelay -v true

I'm not AIX or general OS expert.

My questions are

What AIX use TCP delay for ?
Why we had to change it only on AIX ?
What are pros and cons tcpNoDelay=true and tcpNoDelay=false ?
After changing this parameter to true performance increased.
Are there any other (good or bad) consequences of this ?

Marcin

tillywern

Posted: Tue Mar 13, 2007 9:22 am Post subject: From an engineering perspective.

Centurion

Joined: 28 Jan 2003
Posts: 109
Location: Colorado

From an architect/engineering perspective I would ask do you know what the current configuration of the HTTP listener is. If you do then what does it matter if you change it and run a test. From a practical perspective it would be best to trust the IBM people and run an evaluation.

mqsireportproperties <- get the current value of the configuration parameter.

I'm not trying to be a pain but I would implement their change see if it improves performance and then go back to them and assess risk. If you make the change and it doesn't solve the performance problem then your problem may not be the the product.

I once had spent 6 months telling a client that they had a network issue that was causing an apparent "slow broker". They didn't trust me. In the end the broker was communicating to a database over a 10Mbit network segment that was not supposed to be used.

Are your Windows and AIX systems on the same subnet? Are the paths from the test client to the Broker server the same?

Look here:
http://www-128.ibm.com/developerworks/websphere/library/techarticles/0608_braithwaite/0608_braithwaite.html

tillywern

Posted: Tue Mar 13, 2007 9:28 am Post subject: Sorry

Centurion

Joined: 28 Jan 2003
Posts: 109
Location: Colorado

You got it.

Remember that this value only configures the HTTPListener. So at the OS lever you don't really have to worry about much. Your sending app should manage the stability of the HTTP session so if a failure were to occur then you should be ok.

If you are sending WebService transactions over HTTP and they are updating systems I might opt for a more persistent type connection. But thanks to SOX and Java there should be tons of accounting around any transaction that deals with sensitive data.

marcin.kasinski

Posted: Tue Mar 13, 2007 10:45 am Post subject: Re: Sorry

Sentinel

Joined: 21 Dec 2004
Posts: 850
Location: Poland / Warsaw

Thank you tillywern for your post,

I think you didnt understand my problem.

We did everything what IBM asked us to do and now we are happy with current performance.

Now I would like to know why this parameter has default value false which in my opinion on AIX generats very poor performance.

Now I would like to understand it.

Understand all aspects of this parameter.

I have to be 100% sure that after changing this parameter I didn change something else.

Why IBM decide to set this parametr do false in defaul configuration.

I hope this post shows my fears.

Marcin

tillywern

Posted: Tue Mar 13, 2007 11:41 am Post subject:

Centurion

Joined: 28 Jan 2003
Posts: 109
Location: Colorado

Right. I understood but only right after my first post. I know what you are getting at. IBM sets a stupid default value and then askes you to change it when you bring up the issue. You are just trying to understand why.

Note the free form nature of the command. The -n option is for the name of the config parameter and the -v for the value. I would interpret this to indicate that it is difficult to know the canonical list of possible -n's.

My hypothesis is supported by the lack of documentation on the different values that are acted upon and what valid -v are acceptable.

There are a lot of different tricks the developers use to help resolve problems that are not slated for support in a particular way so they expose them through esoteric methods such as this.

I know I'm not of much help. From a risk perspective, I would think that this fix would be considered to be in scope for your support agreement. If you are fearful that it is not then specifically get it in writing from IBM. That way you are covered from damages or can look to another vendor.

marcin.kasinski

Posted: Tue Mar 13, 2007 11:45 am Post subject:

Sentinel

Joined: 21 Dec 2004
Posts: 850
Location: Poland / Warsaw

tillywern wrote:

OK.

Thanks for your help.

Marcin

Display posts from previous:

Page 1 of 1

MQSeries.net Forum Index » WebSphere Message Broker (ACE) Support » Broker on AIX and TCP delay

Jump to:

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

Protected by Anti-Spam ACP