[zeromq-dev] Some real perfomance numbers from real workloads

[Christian Martinez]

Thank you!

--CM

From: zeromq-dev-bounces at lists.zeromq.org [mailto:zeromq-dev-bounces at lists.zeromq.org] On Behalf Of Andrew Hume
Sent: Tuesday, April 03, 2012 10:28 AM
To: ZeroMQ development list
Subject: Re: [zeromq-dev] Some real perfomance numbers from real workloads

under testing at 1/5 full load, my system stats includes this table for a hot hour:


Summary:         total              ave rate
inbound     33.24M 170.5GB     9.233K/s 47.37MB/s
outbound     106.0M 6.401GB     29.45K/s 1.778MB/s
internal     3.321G 555.2GB     922.6K/s 154.2MB/s

for your purposes, you should add all three lines together.
this is distributed across 8 servers. mostly we
do PUSH/PULL, but for our PUB/SUB, the fanout is of order 150 or of order 8.

            andrew

On Apr 3, 2012, at 11:10 AM, Christian Martinez wrote:


No one has any numbers they want to share?

--CM

From: zeromq-dev-bounces at lists.zeromq.org<mailto:zeromq-dev-bounces at lists.zeromq.org> [mailto:zeromq-dev-bounces at lists.zeromq.org]<mailto:[mailto:zeromq-dev-bounces at lists.zeromq.org]> On Behalf Of Christian Martinez
Sent: Friday, March 30, 2012 7:04 PM
To: zeromq-dev at lists.zeromq.org<mailto:zeromq-dev at lists.zeromq.org>
Subject: [zeromq-dev] Some real perfomance numbers from real workloads

I'd be interested in hearing some of the numbers the community has gotten for its real workloads. Right now I work exclusively with customers leveraging 0MQ on Windows and under real workloads there we've experienced anywhere from 300K to 1.5 million messages a second throughput (the 1.5 million was for 40 byte messages). Latency has always been sub ½ second and for the folks I've worked with that's good enough so I have not spent too much time measuring nanosecond latency on Windows.

I'd love to hear from Linux folks or even other Windows folks what their numbers look like as well as folks working with the various language bindings... Basically what I'm trying to do is get some idea of what the expectation is from the community when they utilize 0MQ. Of particular interest is anyone that has a large number of concurrent subscribers. I haven't done any 0mq work yet with say 500 concurrent subscribers.

I'm not really interested in test harnesses or bench suites etc. I'm looking for actual workload #'s.  Thanks!.


--CM

_______________________________________________
zeromq-dev mailing list
zeromq-dev at lists.zeromq.org<mailto:zeromq-dev at lists.zeromq.org>
http://lists.zeromq.org/mailman/listinfo/zeromq-dev


------------------
Andrew Hume  (best -> Telework) +1 623-551-2845
andrew at research.att.com<mailto:andrew at research.att.com>  (Work) +1 973-236-2014
AT&T Labs - Research; member of USENIX and LOPSA




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20120403/b1c6cb6d/attachment.htm>