[erlang-questions] Story of one performance bug (active, once vs active, N)

[Chandru]

On 13 May 2016 at 09:05, Max Lapshin <max.lapshin@REDACTED> wrote:

>
> Hi.  Will try to share our experience with hunting one strange performance
> bug.
>
>
>
> <snip>


> What is unclear for me:  why was  active,N  better than active,once?   I
> haven't checked but I think that still average size of tcp message and
> amount of messages was the same as in  active,once case.
>
>
>
We had a similar problem with one of our systems at bet365. And the same
workaround also seemed to work. Ultimately though it was because of a
kernel setting which when adjusted made the problem go away with {active,
once}. Strangely enough, pretty much the same day we resolved the issue
Cloudflare wrote a blog post about the exact same problem they faced and
the solution! I hope this is of some use to you.

https://blog.cloudflare.com/the-story-of-one-latency-spike/

cheers,
Chandru
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20160513/b714f422/attachment.htm>