[rabbitmq-discuss] Looking for guidance on R14B04 vs. R13B03 performance

[Matt Pietrek]

Ah, thanks so much for this reply Matthew. It's extremely enlightening and
seems to explain what we're seeing.

A follow up question just for my understanding: Are these "windows" dynamic
in nature (e.g., varying length, varying frequency, somehow tied to current
load)? Or are they more static (e.g. "Once every 20 milliseconds").

Thanks again,

Matt

On Mon, Feb 20, 2012 at 3:28 AM, Matthew Sackman <matthew at rabbitmq.com>wrote:

> On Fri, Feb 17, 2012 at 11:17:24AM -0800, Matt Pietrek wrote:
> > While I'm very happy with these results, I'm still puzzled. Given that
> all
> > clients are writing to the same queue, I'd expect resource
> > contention/blocking in the broker to slow things down much faster than
> they
> > do. Any comment on this? Am I overlooking anything critical to my
> > understanding?
>
> There's a "window" of time during which no fsyncs will occur, and
> txn-publishes will merely accumulate. Only when that window closes will
> the fsync be performed. With a single publisher, each window can only
> have a single publisher's msgs in it. But with many publishers, the
> fsync is effectively amortized, and thus a higher rate is achieved - if
> you monitor with strace or similar, you'll probably see the fsync rate
> is not much different, regardless of the number of publishers you have
> publishing to the same queue.
>
> Matthew
> _______________________________________________
> rabbitmq-discuss mailing list
> rabbitmq-discuss at lists.rabbitmq.com
> https://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20120220/4006001c/attachment.htm>