Mailinglist Archive: opensuse (2348 mails)

< Previous Next >
Re: [opensuse] Software IDE RAID versus hardware IDE RAID
  • From: Sam Clemens <clemens.sam1@xxxxxxxxx>
  • Date: Wed, 02 Apr 2008 18:24:13 -0400
  • Message-id: <47F4078D.1050102@xxxxxxxxx>
Greg Freemyer wrote:
On Wed, Apr 2, 2008 at 5:26 PM, Sam Clemens <clemens.sam1@xxxxxxxxx> wrote:
John Bown wrote:

Hello everyone. I have an old server with an IDE RAID card in it on
which I'd like to install openSUSE 10.3. The problem is, support for
said RAID card (a Dell CERC ATA/100) has been discontinued for some time
now. With that I ask, how feasible and/or advisable would it be to
attempt the following?

1) Install openSUSE and configure it to use software RAID (two mirrored
IDE drives (master/slave) for system, two stripped IDE drives
(master/slave) for data)
2) Enable Encrypted File System (EFS)

Basically, I'm worried that an IDE based machine will be painfully slow
due to the high disk activity. Ideally I would use the machine's
existing SCSI U320 interface, but the required hard drives are just too
expensive.

Get some modern SATA disks.
They're doing I/O at 300 Gbyte/s now.
While they don't have out-of-order queueing like SCSI
and SAS, they do offer high burst speeds for I/O (Serial
Attached SCSI...same cables but use SAS cards -- which
conveniently can ALSO control SATA disks, too. Each port
individually determines if that cable is attached to a
SATA or an SAS disk.

A few misconceptions there:

1) At least for now a single drive does not have the ability to
saturate a 150 Gbyte/sec connection. You only need 300 MB/sec if your
using PMP to multiplex multiple drives on one cable. And then you
need to be using PCI express because PCI is also too slow to
effectively use 300 MB/sec.

I'm not saying that he needs it...only that it's available.


2) SATA-2 drives do now support out of order queueing (NCQ or TCQ,
I've forgotten). But the benchmarks are showing that the Linux Kernel
elevators work so well that little (or nothing) is gained from letting
the drive empty cache out of order. Apparently there are some
specific workloads where NCQ is a win.

3) Strangely, most drives do tie 300 MB/sec to the Sata-2 functions
set, so if you use the 150 MB/sec throttle jumper to slow down the
drive, you loose the NCQ function.

Why deliberately throttle down a drive?
They're slow enough already.
And the SAS/SATA negotiation standard specifies that the
controller and the disk are supposed to negotiate their
speed.


4) Ignoring SAS for a second: I assume the old MB under discussion
does not have any PCIe slots, so I don't think you will find a sata-2
controller for PCI, so I think you are stuck with Sata-1 functionality
and 150MB/sec speed. I don't know about SAS controllers. I have not
researched them.

Go to Adaptec's site. There's a lot of whitepapers on
the subject...and here's a cool presentation by HP
on PHY and signal layer:

www.scsita.org/aboutscsi/sas/tutorials/SAS_Phy_layer.pdf
--
To unsubscribe, e-mail: opensuse+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: opensuse+help@xxxxxxxxxxxx

< Previous Next >
Follow Ups