RE: Does XFS on hardware RAID5 have perfomance issues?

[Murthy Kambhampaty <Murthy.Kambhampaty@goeci.com>]

> -----Original Message-----
> From: 
> Sent: Thursday, March 07, 2002 15:22
> To: Murthy Kambhampaty
> Cc: linux-xfs@oss.sgi.com
> Subject: RE: Does XFS on hardware RAID5 have perfomance issues?
> 
> 
> On Thu, 2002-03-07 at 14:20, Murthy Kambhampaty wrote:
> > > From: Steve Lord [mailto:lord@sgi.com]
> > > Sent: Thursday, March 07, 2002 14:49
> > ...
> > > Subject: Re: Does XFS on hardware RAID5 have perfomance issues?
> > ...
> > > So I do not think it should matter. Let me guess, you do 
> not like the
> > > performance you are getting right now.
> > 
> > We have a SuperMicro S2QR6 motherboard with four PIII Xeon 
> 500 MHz/2Mb cache
> > and 4gb RAM (call it comp1). I get kernel compile times 
> that are thrice as
> > long as on a dual PIII 550 Mhz Katmai processor with 256 MB 
> of RAM (call it
> > comp2); importing a file into MySQL seems to take almost 
> 50% longer on comp1
> > than on comp2. So, I'm chasing down all the bottlenecks and 
> trying to
> > eliminate them. I thought the 2.4 kernel scaled well to 
> sixteen processors,
> > and there is no indication that the highmem configuration 
> slows the kernel
> > down this much, but I might have to get on the kernel 
> mailing list with
> > this.
> > 
> > Thanks for the response to my original question,
> > 	Murthy
> 
> Is the disk setup the same in both - raid5 does have a cost.
> 
The better comparison in this regard is to the backup database server, which
is a dual PIII Xeon 500 MHz/1MB cache (Dell WS620), with hardware RAID5 on a
3ware 7850 controller, with 2 Gb or RAM. I find that kernel compile time and
MySQL import are faster on the dual-proc. mach. than on the quad-proc. mach.
Of course, the quad-proc. server has the benefit of being able to support
many disks, therefore giving the capacity we need. But it is getting a
little uncomfortable that a DP machine with IDE raid is far faster than the
quad-processor SCSI raid box.

Thanks,
	Murthy