nginx high memory usage in benchmarks

merlin corey merlincorey at dc949.org
Wed Jun 24 08:10:02 MSD 2009


Also important to note when doing these benchmarks is whether or not
you ran the benchmark programs on DIFFERENT machines or the same
machine as the webservers.  If it's the latter, you will get noise as
the benchmark programs compete with the webserving programs for
resources.

-- Merlin

2009/6/17 Martin Reissner <mreissner at wavecon.de>:
> hello igor, i found the culprit. it was the access log. i'm pretty new
> to nginx and i just commented the access_log option not knowing that i
> had to set it to off to disable it.
> as i was monitoring the whole system's ram usage i also got the access
> log being written to the filesystem cache. turned it off now and results
> look more promising. thanks for answering.
>
> martin
>
> Igor Sysoev wrote:
>> On Wed, Jun 17, 2009 at 07:16:16PM +0200, Martin Reissner wrote:
>>
>>> hello,
>>> i've been running some benchmarks for static content with httperf and
>>> autobench to compare nginx to apache's worker mpm. i used the default
>>> config on apache (2 worker processes, 150 max clients) and set nginx to
>>> use 2 worker processes with epoll and a keepalive_timeout of 15.
>>> during the benchmarks i monitored the systems ressources with sar and
>>> while i expected nginx to outperform apache in every way possible i
>>> noticed that nginx uses way more ram than apache while staying ~1/3rd
>>> behind it in cpu usage.
>>>
>>> the benchmarks request a 10k file repeatedly over multiple persistent
>>> connections and while apache used ~5mb throughout the tests nginx took
>>> ~70mb. can someone explain to me what the cause of this might be. from
>>> what i read its low memory usage is one of the key features of nginx so
>>> i'm pretty sure somethings wrong on my side.
>>
>> What nginx version have you tested ?
>> How many worker_connections did you set ?
>>
>>
>
>
>





More information about the nginx mailing list