Subject: RE: Mirrored Raid 5 responses
From: volz (volz@koyukuk.at.uaa.alaska.edu)
Date: Tue Nov 04 2003 - 17:02:13 AKST
Mike, Mac, Dee and Jim,
Thank you for your responses.
I did not give enough detail.
We have a DELL 755N NAS box with a perc 3DC and an attached 220S. The system is
running Windows powered OS. WE have tried both a single large volume with
dedicated hot spare and two smaller volumes with global hot spare. Both have
gone down in flames with multiple drive failures. It has been an unreliable
system and does not function well as an NFS server. Furthermore, with the help
of DELL level 1 technicians we have hosed the data twice.
So we have pretty good hardware I would just like to put together a more
resilient system that runs an operating system I understand and can give us the
type of performance we paid for.
So to address some of your answers and help me sort out a good plan, I would
like to see if I understand the issues and capabililties of the linux kernel and
lvm.
>From: "Wadell, Jim S (SAIC)" <WadellJS@BP.com>
>To: "W.D. McKinney" <deem@wdm.com>, "volz" <volz@koyukuk.at.uaa.alaska.edu>
>
>Last part first. Raid 4 is similar to raid 5, except all parity is on one disk,
date on the rest. Quick google indicates that Linux software raid does do raid4.
>Downside is that the parity disk gets beat up pretty bad. There are no
additional security (resiliency) implications that I know about. You survive a
single disk.
>
Are you saying that despite the fact that parity is on its own disk only one of
the data disks can go bad at a time? This is how I understand RAID4. So there
would be no gain. Toss that one.
>Again a quick google indicated some work in Kernel.org about raid15 (2001),
doing raid5 out of raid1 pairs. I knod of get the impression that it was an "I
did it because I can" sort of deal.
>
I was considering more of a RAID 51 scenario where if you hose the raid you have
a way to restore in hours versus days of tape restore. I see how RAID15 would be
prefereable because you could rebuild from mirror in a short time. Seems overly
expensive of drive space. Put that one aside for the moment.
>Assuming that you want to survive something other than physical destruction of
the box (always hard because of power, etc), I would lean toward raid5 with some
hot spares. Assuming that you do not loose two disks simultaneously, you just
keep going. I might entertain a three disk raid1 set for the / and /usr /var
stuff. However, when you come up with scenarios in which you are going to loose
more than a single disk, due to just disk failure, then you are envisioning
power (dual or triple power supply box with separate circuit/ups units),
controller failure (1/2 to 1/3 or ? of the drives), etc. Power would be most
likely, controllers less so.
Hopefully the system will be on the NAS box and the raid will be a bunch of
disks for data only. THe system disks would likely be on a separate controller
with RAID1 for the system. The disk enclosure has redundant power supplies with
stout UPS and powerchute network shutdown on the server.
We have had multiple disk failures. A case of a failed disk pulling a neighbor
down I think. I attribute it to Array Manager the disk manger supplied with hhe
system.
My guess is the best thing we could do is do two 5 disk RAID5 volumes with a
hotspare on each. That way the chance of failing two disks at a time are less
tha two out of ten and we don't get the performance hit of ten spindles on a
volume unless data is written to both volumes simultaneously. Pretty much our
current configuration but implemented through linux instead of windows.
>
>Is this what you were looking for?
>
Yes thank you.
- Karl
>Jim
>
>-----Original Message-----
>From: W.D. McKinney [mailto:deem@wdm.com]
>Sent: Tuesday, November 04, 2003 12:56 PM
>To: volz; Wadell, Jim S (SAIC)
>Subject: Re: Mirrored Raid 5
>
>
>On Tue, 2003-11-04 at 12:45, volz wrote:
>> Hello,
>>
>> Does anyone have experience with mirrored raid 5's? Can linux scsi drivers
and
>> logical volume mangement encompass such a thing? Variously called 15, 5+1.
>> Anybody know where that documentation lies? I googled "RAID 5 mirror" and
got
>> some hits but haven't found the substantive info. There is even a newsgroup
>> post by someone using redhat 7.3 and the same hardware I want to use to build
a
>> 1 tbyte RAID 5. but no specifics
>>
>> Also is anyone aware of way(s) to put parity on a disk or mirrored disks and
>> stripe the data w/o parity across an array? Does this provide any additional
>> security benefit. Most importantly, would it be possible to survive the death
of
>> multiple array disks in this case?
>
>My friend Jim Wadell is the person to ask.
>Jim can you help please ?
>
>Dee
>
>>
>> Or more genrally the question is: given 12 scsi disks, a linux kernel, couple
of
>> fast processors and decent scsi card, how much safe storage can I make? (safe
>> --> Can we suffer two disks down?)
>>
>> - Karl Volz
>>
>> Research Associate
>> Alaska Experimental Forecast Facility
>>
>> volz@aeff.uaa.alaska.edu
>>
>> ---------
>> To unsubscribe, send email to <aklug-request@aklug.org>
>> with 'unsubscribe' in the message body.
>
>
>
---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
This archive was generated by hypermail 2a23 : Tue Nov 04 2003 - 17:05:13 AKST