[SATLUG] OT: RAID drives

Brad Knowles brad at shub-internet.org
Thu May 12 01:58:56 CDT 2011


On May 11, 2011, at 9:34 PM, Alan Lesmerises wrote:

> Performance is not a big driver since there is only a few computers that will be accessing the RAID for data backups.

In my experience, performance is usually not a concern -- until it is.

When you have a failure that you need to quickly restore from so you can get that machine back online, that's when it really matters.  That said, what I might consider to be unacceptable low performance might actually be just fine for you, or vice-versa.  So, again we get back to the answer "it depends".

> I should have also mentioned that it is a hardware RAID (Adaptec I2O controller).

I'm not sure that I'm familiar with that particular controller.  Hardware RAID controllers that I have used in the past have certainly had their advantages when it came to working in degraded mode -- they aren't as fast as software RAID when everything works well, but when it all goes down the toilet they usually hold up a lot better because that's one fewer thing that is trying to grab all the CPU it can on the main processor.

OTOH, hardware controllers that I have used in the past have also been more finicky about the drives that they will let you configure and add to the same raidset.  So, both advantages and disadvantages there, but it all depends on the particular controller you have and how much really is done in hardware and how much really is done in software and how that code is written.

> The drives I have now are Seagate Cheetah Ultra 320 SCSI ST3146807LCVs, and the ones I'm looking at are Seagate Cheetah Ultra 320 ST3146807LCs.  The specs look essentially identical, but if like you said -- sometimes the OEMs re-use the same model numbers but with slight changes -- the differences could be enough to matter.

Although I did an internship at Imprimis/Seagate back in the late 80s, I haven't kept up with any of my contacts there and I don't know anything special about their more modern drives.  It is entirely possible that these drives might be similar enough that you won't notice the difference, or they could be different enough that they simply won't work at all -- or anywhere in between.

If you can find someone who has deep knowledge of the specific mechanisms in question and the controller you're using, you may be able to get an answer before you decide whether or not you have to just jump in and hope that you can learn to swim before you sink.


For my own part, I would be strongly tempted to buy enough of the new drives to completely replace all the old ones (whether or not they have failed yet), copy all the data over from the old raidset to the new one (which might even be on a different machine), and then retire the old raidset as a unit.  In my experience, what you tend to see with drive mechanisms is what we call a "bathtub curve", where they experience a higher rate of failure early in life, then a lower rate of failure for most of their operational life, and then as they get older they will start failing at or near the same time as other drives of the same make/model that were manufactured around the same time.

So, even if the other drives are still going strong for the moment, you've already had your "canary in the coal mine" event with the first failure, and it might be a good idea to consider replacing the whole array at once, which you will hopefully be able to complete before you get any more failures.

> Thank you for such a thorough consideration of my question (you must have started typing as soon as my original message went out!).

Pretty soon afterwards, yes.  ;-)

--
Brad Knowles <brad at shub-internet.org>
LinkedIn Profile: <http://tinyurl.com/y8kpxu>



More information about the SATLUG mailing list