[zeromq-dev] disabling epoll()

[Pieter Hintjens]

I'd happily do it in C as a proof of concept if you like. I'm not up to
changing the current poller class.
On 4 Oct 2010 15:39, "Martin Lucina" <mato at kotelna.sk> wrote:
> ph at imatix.com said:
>> On Mon, Oct 4, 2010 at 3:23 PM, Martin Lucina <mato at kotelna.sk> wrote:
>>
>> >> Could you explain how the build currently decides to use e/poll() or
select()?
>> >
>> > This is determined at compile time by configure which sets the relevant
>> > ZMQ_HAVE_XXX to detect the OS and the rest is done in the ifdefs in
>> > poller.hpp.
>>
>> Ah, so it's purely system dependent, with manual override...
>
> Yes.
>
>> > Where's the patch? It's not in your original email, maybe you forgot to
>> > attach it?
>>
>> Goshdarnit, I did forget! And then I pulled a new snapshot of zeromq
>> and lost it. Anyhow it was just a line in configure.in, near the end:
>
> I found it, in another thread, sorry...
>
>> AC_CHECK_FUNCS(select poll epoll kqueue)
>>
>> Since there is no actual problem to solve - as you said,
>> cross-compiling for a 2.4 kernel needs manual override - it's moot.
>
> Precisely.
>
>> I guess we don't want to try stuff like switching between select() and
>> e/poll() depending on the number of sockets we're working with?
>> There's some evidence select() is faster for smaller sets[1].
>
> That's an interesting idea; I'm curious to see how you implement it.
>
> :-)
>
> -mato
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20101004/6c69810c/attachment.htm>