-
1. Re: Using Stomp::Receipt - Performance is slow
clebert.suconic Sep 6, 2011 9:35 PM (in response to ameyamk)What version?
-
2. Re: Using Stomp::Receipt - Performance is slow
ameyamk Sep 7, 2011 1:10 AM (in response to clebert.suconic)2.2.5 Final
-
3. Re: Using Stomp::Receipt - Performance is slow
ataylor Sep 7, 2011 3:48 AM (in response to ameyamk)have you tried another client to see if the performance issue is client side?
-
4. Re: Using Stomp::Receipt - Performance is slow
ameyamk Sep 7, 2011 12:16 PM (in response to ataylor)I tried using Stomp Client from different machines, using perl and Ruby client. I have not tried Java client yet. But I see similar performance from both the stomp clients from different machines.
-
5. Re: Using Stomp::Receipt - Performance is slow
clebert.suconic Sep 7, 2011 12:24 PM (in response to ameyamk)Are you measuring a single producer with persistent messages?
on that case you need to serialize each message send to wait the server to kick persistence. Multiple producers will use the throughput available at the server.
Isn't that the case that's happening to you?
-
6. Re: Using Stomp::Receipt - Performance is slow
ameyamk Sep 7, 2011 1:52 PM (in response to clebert.suconic)I have 10 different producers from 3 different machines, and a single consumer. I am measuring performance at consumers end, which gives me sort of aggregate performance for all 10 producers, as consumer is much faster than the producers.
I was hoping that single producer might be slow as its esentially blocking on the ack from broker, however total broker capacity would not affect significantly. If I send messages without receipt, broker is able to handle upto ~5000/messages per second, but in receipt mode this is down to ~200-300 messages persecond.
-
7. Re: Using Stomp::Receipt - Performance is slow
gaohoward Sep 8, 2011 12:54 AM (in response to ameyamk)Hi ameya,
Is your consumer located in a different machine than the hornetQ server or at the same machine?
Howard
-
8. Re: Using Stomp::Receipt - Performance is slow
ataylor Sep 8, 2011 5:33 AM (in response to ameyamk)I have 10 different producers from 3 different machines, and a single consumer. I am measuring performance at consumers end, which gives me sort of aggregate performance for all 10 producers, as consumer is much faster than the producers.
Thats not strictly true, this does not give you the aggregate perf of the producers, it gives you the performance at which a single consumer can run. remember scalability comes with multiple producers/consumers, however since you think the problem is in the send its probably an ok assumption.
I see that this client is a stomp extension and uses some transactionality, i can only assume that this is all done in the client which may be where the extra over head is. If you can provide an *easy* runnable test i will take a look when i get some time