Re: [PATCH] af_unix: unix_write_space() use keyed wakeups

[Eric Dumazet]

Le mercredi 24 novembre 2010 Ã 00:20 +0000, Alban Crequy a Ãcrit :
> Le Wed, 24 Nov 2010 01:27:56 +0200,
> Pauli Nieminen <pauli.nieminen@xxxxxxxxxxxxxxx> a Ãcrit :
> 
> > ----- Original message -----
> > > Le Sat, 30 Oct 2010 08:44:44 +0200,
> > > Eric Dumazet <eric.dumazet@xxxxxxxxx> a Ãcrit :
> > > 
> > > > We still loop on 800 items, on each
> > > > wake_up_interruptible_sync_poll() call, so maybe we want to
> > > > optimize this later, adding a global key, ORing all items keys. I
> > > > dont think its worth the added complexity, given the biased usage
> > > > of your program (800 'listeners' to one event). Is it a real life
> > > > scenario ?
> > > 
> > > Pauli Nieminen told me about his performance problem in select() so
> > > I wrote the test program but I don't know what exactly is the real
> > > life scenario.
> > > 
> > 
> > Real world scenario is xsever that has tens client connections to
> > manage. When xserver is happily sleeping at seclect call some client
> > reading events/replies from server triggers in kernel looping over
> > all xserver fds. xserver isn't waiting for socket to became writeable
> > in ussual cases so kernel schedules back to client.
> 
> But are they SOCK_STREAM or SOCK_DGRAM sockets? The patches fix
> performances with SOCK_DGRAM. If the xserver scenario is with
> SOCK_STREAM sockets, your problem is probably still unfixed.
> 


It should not matter ?

commit 67426b756c4d52c51 (af_unix: use keyed wakeups ) makes
unix_write_space() call wake_up_interruptible_sync_poll() instead of
wake_up_interruptible_sync().

So it should be fixed for both STREAM/DGRAM sockets ?



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