[rabbitmq-discuss] RabbitMQ throughput

[Suhail Doshi]

On Thu, Oct 1, 2009 at 1:32 PM, Matthias Radestock <matthias at lshift.net>wrote:

> Suhail,
>
> Suhail Doshi wrote:
>
>> Does anyone know what rabbitmq throughput is from items sitting in the
>> queue to sending them to a consumer?
>>
>> Lets say we're doing 200 items in the queue a second with 50 consumers
>> waiting to process the item, usually they do in < 30ms.
>>
>
> So that's ~7kHz, which doesn't look unreasonable, but there are many
> factors at play here which we'd need to know about in order to make a better
> assessment:
>
> - are these messages persistent or transient? Persistent

- how big are the messages? Less than 500 bytes usually
> - do the consumer consume in no-ack mode, or ack mode? ack mode
> - are the consumers sharing a connection (or even channel) or do they all
> have their separate connection/channel? do not share a channel/connection
> - where are the consumers in the network topology in relation to the
> server? very close, same rack presumably
> - what's the hardware spec of the server machine? quad core, 4 GB of RAM
> - what version of Erlang are you running? 5.7.2
> - what AMQP client are you using? Python library created by Barry Pederson
>
> Also, ...
>
>  Watching our consumer logs, it feels like they don't get items fast
>> enough from rabbitmq, is that possible or is it me?
>>
>
> Can you instrument the consumers to measure how long they are blocked
> waiting for messages from RabbitMQ? Yes possibly, I should check you're
> right.
>
> Also, with 200 messages shared across 50 consumers each consumer will only
> get four items, which is a rather small figure to do any meaningful
> statistical analysis with.
>
>
> Regards,
>
> Matthias.
>



-- 
http://mixpanel.com
Blog: http://blog.mixpanel.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20091001/c60b6a4d/attachment.htm