Re: Scientific Linux Workstation

From: John-Paul Stewart (jpstewart_at_binaryfoundry.ca)
Date: 06/30/04 

Date: Wed, 30 Jun 2004 17:46:15 -0400

Gary J. Weiner wrote:
> Hello my brothers and sisters out in Penguinland.
>
> I have been tasked with building a Linux workstation for scientific and
> engineering applications, mostly running under Matlab. The applications
> are for particle accelerator calculations.
[snip]
> The ASUS P4C800-E looks like a nice board to go with this chip.

You may want to seriously consider a dual CPU setup. I don't know if
Matlab supports it for sure, but I'd be quite surprised if they *don't*
support multiple CPUs.

[snip]
> STORAGE
>
> I was orginally going to go with RAID 1 (mirror) with 2 WD Raptors run
> off the SATA controller on the MOBO. But it seems there are issues with
> SATA and even more issues with the on board RAID controllers.
>
> So maybe I'll go with tried and true SCSI, two Seagate 73GB 10,000RPM
> SCSI Hard Drives and an Adaptec Ultra320 SCSI Card, Model "29320A-R" for
> RAID 1

If you're going to go SCSI, make sure you get a motherboard with 64-bit
PCI slots (or better yet, PCI-X). Even Ultra160 can overwhelm a 32-bit,
33MHz PCI slot. If you're pushing around serious quantities of data,
that can be quite a bottleneck. (Note that even built-in SCSI is
connected to the PCI bus even if it isn't a card in a PCI slot.)

For example, I've got one motherboard here with dual U160 and Gigabit
Ethernet all on the same 32-bit 33MHz PCI bus and bus contention is a
*huge* problem. RAID speed (any RAID level) is the same as single disk
speed strongly suggesting bus contention issues. Another system with
PCI-X provides dramatically better disk performance---even with fewer
disks to stripe across.

Relevant Pages

  • Re: Bad performance
    ... I suggest you take the time to benchmark the stripe size as benefits are ... incremental compile without any changes and no actual compilation occuring - ... The raid I use is the stock Intel ICH5R SATA raid available on Intel 875 ... other raid controllers that live on the PCI bus so are limited to PCI ...
    (microsoft.public.windows.server.setup)
  • Re: Live TV Hangs
    ... What is your disk drive configuration? ... tuner card is in a PCI slot next to an AGP slot try moving it. ... GIG Ram, 500 Gig HD ...
    (microsoft.public.windows.mediacenter)
  • Re: RAID - speed limitation???
    ... > If I test each controllers separately I get a speed of around 179 MB/s ... If the two RAID controllers are both on ... the same PCI bus, that will certainly be a bottleneck. ...
    (comp.os.linux.hardware)
  • Re: Linux: Why software RAID?
    ... Generally, you want software RAID unless your PCI bus (or more rarely, ... your CPU) is getting saturated. ...
    (Linux-Kernel)
  • Re: SATA RAID with Promise TX4 (Was: [opensuse] Recommendation for sata card?)
    ... Bonnie++ showed that the RAID was in the same ballpark as a PATA drive ... I didn't think that PCI bus would be the limiting factor. ...
    (SuSE)