> From: Zach Brown <zab@zabbo.net>
> Subject: Re: Got benchmarks?
>
> > ... But what about the kernel fixes? Didn't you use a wake-one scheduler
> > patch? ...
>
> it didn't make a darn bit of difference, which makes sense when you think
> about the work load. there are enough concurrent connections that we
> spend a far greater amount of time scheduling between the apaches handling
> connections than we do having apaches all waking on accept. the path to
> serve a connection is 20someodd syscalls. and while thats going on we
> have interrupts arriving. we end up shceduling all over the place.
>
> notice that the run queue is always proporional to the number of
> connections, _not_ the number of listening apaches.
>
> and in fact, andrea's patches made performance drop off worse than it did
> under stock 2.2.10.
>
> -- zach
Maybe that patch someone posted to handle really long runqueues
well would be useful after all.
- Dan
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/