Stress Test Question

Jan 6, 2012 at 4:21 AM

I modified the stress test sample to publish individual messages rather than batch (Topic.PublishMessage(StressTestMessage.New) and altered the batch size to 1 and it seems like the subscriber will stop receiving messages after a couple hundred but I have been unable to identify a cause.  Have you stress tested in this type of configuration? 

Coordinator
Jan 6, 2012 at 4:27 AM
Edited Jan 6, 2012 at 7:01 AM

The problem i believe is the 127.0.0.1 address. Try using replacing the ip address property with the ip address that shows up for IPv4 and you should notice a difference. I will update the stress test sample with a solution that retrieves the IPv4 address. I tried the configuration and the only thing that blocks the subscriber from getting more messages is the TCP TIME_WAIT which currently does not allow you to reuse port for each new connection. I am thinking of updating the TCP Transport to cache the TcpClient in order to increase throughput for sending and which in turn would remove the problem with been able to continue to receive messages. Until I fix it, If you have a fix for it, it is very much welcome.

Jan 6, 2012 at 8:29 PM

Made the change you suggested but didn't see any change in behavior.  What I did notice was that these large batches will eventually complete but subcribers will see long delays.  I ran a sample of 500 messages and though the program initially stopped at 427 after only a few seconds, the remaining messages did complete after about 30 minutes. 

Coordinator
Jan 6, 2012 at 9:25 PM

The problem was tcp time wait. There is a lag of 60 seconds for time wait. The reason you saw the message after is because the retry kicked in after there were enough port available. You should notice that it takes longer if you run multiple instance of the test. the transmission of messages are really fast when moving between the redis nodes and that is why the port are running out so fast

On Jan 6, 2012 3:29 PM, "dsmith1569" <notifications@codeplex.com> wrote:

From: dsmith1569

Made the change you suggested but didn't see any change in behavior. What I did notice was that these large batches will eventually complete but subcribers will see long delays. I ran a sample of 500 messages and though the program initially stopped at 427 after only a few seconds, the remaining messages did complete after about 30 minutes.

Read the full discussion online.

To add a post to this discussion, reply to this email (pServiceBus@discussions.codeplex.com)

To start a new discussion for this project, email pServiceBus@discussions.codeplex.com

You are receiving this email because you subscribed to this discussion on CodePlex. You can unsubscribe or change your settings on codePlex.com.

Please note: Images and attachments will be removed from emails. Any posts to this discussion will also be available online at codeplex.com