[rabbitmq-discuss] Subscription Performance

Marcos Torres marcos.torres at gmail.com
Thu Oct 31 13:33:58 GMT 2013


Simon,

I'm using the .NET library, specifically the EventingConsumer. I'm not
setting the prefetch count, I will look into that.
Also, I will start doing some performance with PerfTest as suggested.

Regarding Consumers, we have 1 per queue, that's because we handle a lot
(700) of queues. If increasing the Consumer count will have direct impact
in the delivery, I can explore into that but I cannot match the Producer
count, that's for sure.

Thank you.


On Thu, Oct 31, 2013 at 9:54 AM, Simon MacMullen <simon at rabbitmq.com> wrote:

> On 31/10/13 12:48, Marcos Torres wrote:
>
>> We're publising messages with a rate between 1000 msg/sec and 2000
>> msg/sec in our dev machine. This is OK for us in our current dev status.
>>
>> The problem is with the delivery/ack speed. We're getting at most 150
>> msg/sec with all our logic disabled : this means we are just
>> deserializing and ack the message.
>> This is creating a bottleneck.
>>
>> I was excepting similar numers to publshing, now I wonder if my
>> assumption is correct.
>>
>
> You should be able to get approximately similar performance at the
> consumer end as you get on the producer end.
>
> So with that in mind:
>
> * Are you setting prefetch count (basic.qos)?
> * Are you using a different client library?
> * How many consumers are you attaching to a queue?
> * What performance do you get consuming the messages with PerfTest?
>
> $ runjava.sh com.rabbitmq.examples.PerfTest -h amqp://user:pass@server-x0 -p -u <queue-name>
>
> Cheers, Simon
>
> --
> Simon MacMullen
> RabbitMQ, Pivotal
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20131031/97cd2eb1/attachment.htm>


More information about the rabbitmq-discuss mailing list