Re: [squid-users] To RAID or not to RAID...

If we are talking RAID, it means talking hardware. It seems almost
impossible get a good performance by using UltraATA for RAID although it is
UltraATA100. I tried some ATA RAID controller and get a very disappointed
result.

In case, we use ATA RAID that software based (by OS, such as NT),
performance goes much worse. Because it consumes the main CPU resources.

SCSI has feature to simultaneously accessing SCSI devices at the same time
that ATA does not have this feature. So, UltraSCSI160 (data rate 160 MBps ?)
+ RAID card + RAM + battery backup (and configure it as RAID 5) should be
the best choice. It is slower than disk stripping but it's parity will
protect the HDD system and faster than disk mirroring (or single disk).

AFAIK, disk mirroring has the worst performance than other RAID type.

Awie

----- Original Message -----
From: "Henrik Nordstrom" <hno@marasystems.com>
To: "Winston Gutkowski" <winston.gutkowski@eztext.com>; "Squid-Users"
<squid-users@squid-cache.org>
Sent: Wednesday, June 05, 2002 2:48 PM
Subject: Re: [squid-users] To RAID or not to RAID...

> On Wednesday 05 June 2002 01:51, Winston Gutkowski wrote:
> > Thanks Henryk. Most informative.
> > I'm interested in your figure of 3/5 for seek time on RAID 1
> > though. Is this based on IDE or SCSI, software or hardware RAID?
>
> No, simply RAID 1.
>
> When you do RAID 1, all drives in the mirror (usually two) will be
> involved on each write, and one of the drives will be involved on
> read, almost degrading the number of spindles to the half compared to
> using each drive individually.
>
> If you have a crappy I/O system then the situation will always be
> bad.. RAID or not..
>
> > When I was at Sun I had to configure quite a few servers and we
> > were told that, even with software RAID, a RAID 1 disk would
> > generally outperform a single disk on reads and run @ 60-70% of a
> > single disk on writes.
>
> True, but two disks used in parallell gives you close to 200%
> performance on both reads and writes, and considering that a Squid
> workload consists of mainly small writes..
>
> Also note that the SUN figures is based on a typical I/O workload
> where quite many I/O operations are large, while the Squid I/O
> workload is far from typical, with lots of small I/O randomly spread
> out everywhere..
>
> > PS: Really like the ability to shrink or expand a cache just by
> > adding cache_dirs. Very nice. Do you have to stop and start squid
> > to activate?
>
> You will need to restart Squid when you delete a cache_dir. Adding can
> be done at any time.
>
> Regards
> Henrik
Received on Wed Jun 05 2002 - 02:08:54 MDT