[nycbug-talk] RAID controllers - what doesn't suck?
Mon Sep 13 18:25:27 EDT 2004
On Mon, 13 Sep 2004 alex at pilosoft.com wrote:
> Its a bad idea. If controller says drive is bad, its a hint - CHANGE THE
> DRIVE *BEFORE* IT FAILS. Controller may mark drives as bad for many
> reasons - such as SMART diagnostics even before failure - and that's a
> good thing ;)
That's what I thought the first few times. :) One box that I no longer
care for had around a dozen IBM drives in it, and under my care it ate
about 4 of those. Since then I think it ate another four. I believe that
the next person to work on it spent a good deal of time and ended up
flashing the firmware on the remainder, which seems to have fixed it.
The box in question has gone through three out of four. So I guess I'm
mainly suspicious of the drives, but not necessarily in the sense that
it's a physical failure.
> Does this happen with all IBM drives?
So far, yes. Another employer with about 8 zero-channel adaptec raid
controllers has had much better "luck", but I believe the bulk of those
drives are Seagate.
>> One other question... I'm going to be setting up a new shell server
>> that also serves up a decent amount of ~user webpages. That too will
>> need a RAID controller. I'm not really up on what advantages SATA has
>> over normal IDE drives, but I wonder if perhaps SATA RAID would be
>> sufficient in this case? Lots and lots of random reads, not very
>> sequential. Opinions?
> That depends. :)
> SATA's main advantage is better wiring, really. Speedwise, disks are not
> close to reaching limit of ATA-100 anyway. Of course, 10K RPM SATA disk is
> going to be better - but they are just as expensive as 10K SCSI drives...
Thanks, I'm really in the dark about SATA. I need to read up on that
somewhere (my G5 has one). We're not talking about a large sustained
throughput on the web/shell box, but a decent amount of tiny, random
reads, which I assume SCSI is better at since it has command
queueing/tagging. Although maybe ATA drives now offer similar features -
it's been awhile since I adopted my "SCSI on all servers" stance.
More information about the talk