Maximum performance for caching

[Ingo Molnar]

* Batara Kesuma <bkesuma at ml.gaijinweb.com> wrote:

> Hi William,
> 
> > What kind of hardware do you use? (proccessors, RAM, SCSI|IDE).  What
> > kernel are you running?
> 
> My TUX servers are running:
> tux-3.2.18-1
> kernel-smp-2.6.5-1.358
> 
> The hardware is:
> Pentium 4 2.80GHz (with HT)
> 1GB RAM
> 80GB HDD SATA 
> 
> But mostly of the images are served from NFS. The NFS server has the
> same specs, except it has 250GB HDD SATA with RAID 1. I think the
> bottleneck is on the NFS server.

what is the typical 'hot content' size (that users are accessing) - much
larger than 1 GB?

> > Looking at the statistics in the apache access_log, almost 70% of the
> > requests are static content, so tux would be a fantastic thing here. 
> > TUX is really brilliant with static content.
> 
> Yes, more than 80% of the requests are static contents. TUX is
> amazing, but the problem is the NFS server :(

you can increase IO parallelism by increasing NR_IO_THREADS from 32 to
64 (or higher) in include/net/tux.h. (Note that this will add a whole
bunch of IO threads but with a high-latency NFS server those are needed
if your working set is much larger than the 1GB of RAM.)

could you run 'top', enter 'h' and 's60<enter>' (show all threads and 60
second measurement interval) and copy & paste the full screen (once the
first 1 minute measurement has been displayed) and send it to me?

'D' state processes and a high load are not necessarily an issue unless
the response times of your server are bad and still you have alot of
idle CPU time and idle network bandwidth.

	Ingo