Re: Slowdown on high-load machines with 3000 sockets

[Christian Schmid]

No i am only using 4 tasks with Poll-API and non-blocking sockets. Every socket gets a 1 MB 
read-ahead. This are 4000 MB Max on a 8 GB machine.... Shouldnt thrash.

Rik van Riel wrote:
> On Sun, 27 Feb 2005, Christian Schmid wrote:
>
>
> > > Christian, how big are the data blocks you sys_readahead, and
> > > how many do you have outstanding at a time ?
> >
> > I am reading 800000 bytes of data as soon as there is less than 200000 
> > data in the cache. I do assume the cache doesnt kill itself. I have 8 GB 
> > machine with 7,5 GB free for caching.
>
>
> OK, so you read 800kB with 4000 threads, or 3.2GB of
> readahead data.  The inactive list is quite possibly
> smaller than that, at 1/3 of memory or around 2.6 GB.
>
> It looks like the cache might be killing itself, through
> readahead thrashing.
-
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/