> On Wed, Aug 27, 2003 at 11:04:03PM -0700, Bryan Vyhmeister wrote:
>   | I have recently built a single Xeon 2.4 GHz system based on a
Supermicro
>   | X5SS8 motherboard along with an Adaptec 2010S Zero Channel Raid
card. I
>   | am using two Seagate 36 GB, 10,000 RPM, Ultra320 SCSI hard drives
in a
>   | RAID 1 setup. I have 1GB of ECC DDR RAM installed.
> 
> If you have a spare modern-ish IDE drive, try using that and compare
> against the 2010S.
> 
> In my experience, the 2400A and 2100S cards don't have great
> sequential throughput in RAID-1, and I've seen various articles
> indicating that others have noticed that.  If the 2010S is based
> on a similar CPU and firmware as the 2100S, it may have the same
> "lame RAID-1 performance issues".

I have seen this mentioned several times. That is actually why the 2400A
in the P4 is doing RAID-5 instead of RAID-1.

> (To put this in perspective, I was getting ~ 40MB/s sequential on a
> single drive on a 2400A, and when I mirrored that drive it dropped to
> 10MB/s).

What is the best way to test sequential performance?

> That all said, with 1GB of RAM you shouldn't be hitting the disk
> that much, compiles are more IO/s bound than sequential throughput
> bound, and the 2400A/2100S are quite good for IO/s in a RAID-1
> configuration.
> 
> 
>   | Has anyone else observed similar issues on a Xeon system? I am
really
>   | not sure what to think. It seems very strange.
> 
> We've got a build host that's a dual Xeon 2.8GHz with 2GB of RAM and
> a couple of IDE disks software mirrored with RAIDframe, and it is
> very very zippy.

I think something else must be going on here besides just the disks
being slow. I started a snapshot build last night at about 11:30 PM on
the Xeon and it is still going now at 6:15 AM. It has only gotten as far
as the crypto sources and it seems to be going very, very slowly. I
started the same build on the P4 system at about 12:30 AM and it quickly
ran into trouble and quit by around 1:30 AM. I am still waiting for the
Xeon to run into that error. Sources have been checked out from the same
date so the build should also run into an error on the Xeon. I am still
baffled as to what could be causing this system to run so slowly.

Bryan