System profiling tools?

Tue Mar 22 14:19:02 2005

This reminds me, can you specify memory space for a software raid cache?

If, for example, we use the specs mentioned in the previous post, each SATA
drive will burst up to 150Mbs, but only has a sustainable transfer of around
40-45Mbs.  In addition, all drives will be sharing the same bus, and since
RAID 5 has only fair write performance du to overhead.  Thus if I can pump an
effective rate of 100MBs down the pipe to a computer, it is easy to see that
even with a processor fast enough to handle it, the write performance would
drop drastically after only short time.  However, if I could bypass the north
bridge (as you can do with most AMD 64 machines) and drop the incoming data
into a memory space used as a dedicated raid cache, of say 4GB, then your
apparent write rates would increase dramatically.

On Tue, 22 Mar 2005, Alan Dayley wrote:

> Let's suppose that I have a Linux file server.  It is working well and
> does nothing (discounting cron or other system processes) but serve
> shares.
> 
> The server has 1Gb/s ethernet interface to the net and four 150MB/s SATA
> hard drives in Linux software RAID 5.  But, for all this hardware
> bandwidth I only get an effective throughput from my server of 40MB/s data
> transfers.
> 
> I want to find the overhead.  Is it the software RAID?  Is it the SATA
> controller?  Is it the IP stack?  Or a combination of these or other
> things?
> 
> What software tools are available to profile my system and help pinpoint
> the bottle neck(s)?
> 
> Alan
> 
> 
> 
> _______________________________________________
> PLUG-devel mailing list  -  PLUG-devel@lists.PLUG.phoenix.az.us
> http://lists.PLUG.phoenix.az.us/mailman/listinfo/plug-devel
> 