Re: [PATCH] rfc: threaded epoll_wait thundering herd

[Chase Venters <chase.venters@clientec.com>]

On Sat, 5 May 2007, Davide Libenzi wrote:

> On Fri, 4 May 2007, Davi Arnaut wrote:
>
> > Hi,
> >
> > If multiple threads are parked on epoll_wait (on a single epoll fd) and
> > events become available, epoll performs a wake up of all threads of the
> > poll wait list, causing a thundering herd of processes trying to grab
> > the eventpoll lock.
> >
> > This patch addresses this by using exclusive waiters (wake one). Once
> > the exclusive thread finishes transferring it's events, a new thread
> > is woken if there are more events available.
> >
> > Makes sense?
>
> Theorically, make sense. I said theorically because all the use
> epoll_wait MT use cases I've heard of, use a single thread that does the
> epoll_wait, and then dispatch to worker threads. So thundering herd is not
> in the picture. OTOH, it does not hurt either.
> But, that code is completely changed with the new single-pass epoll delivery
> code that is in -mm. So, I'd either wait for that code to go in, or I
> (or you, if you like) can make a patch against -mm.

*raises hand*

I'm working on event handling code for multiple projects right now, and my 
method of calling epoll_wait() is to do so from several threads. I've 
glanced at the epoll code but obviously haven't noticed the wake-all 
behavior... good to know. I suppose I'm going to have to hack around this 
problem by wrapping epoll_wait() calls in a mutex. That sucks - it means 
other threads won't be able to 'get ahead' by preparing their wait before 
it is their turn to dequeue events.

In any case, I think having multiple threads blocking on epoll_wait() is a 
much saner idea than one thread which then passes out events, so I must 
voice my support for fixing this case. Why this is the exception instead 
of the norm is a little baffling, but I've seen so many perverse things in 
multi-threaded code...

> - Davide

Thanks,
Chase
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/