|
|
I'm going to be setting up a PowerVault MD3000 enclsoure with an NF500
PowerEdge server. The enclosure will have 15 x 1TB SATAII drives (7200rpm).
I plan to use one drive as a hotspare but my question/concern is regarding
the RAID configuration.
Should I:
1. split the configuration into two separate RAID5 packs (7 drives - RAID5,
7 drives - RAID5, 1 drive - global hotspare) and thus two volumes.
2. keep it as one large volume (14 drives - RAID5, 1 drive - hotspare).
My preference is option #2 as I will be using this storage purely for user
home directories so it would be much simpler from a management point of view
(configuration of home path in Active Directory -
\\servername\share\username). But my concern with option #2 is the rebuild
time in the instance of a drive failure.
Has anyone had any experience in this area that can offer an opinion?
Thanks.
*note: will be using GPT to span the 2 TB volumes.
*note 2: will be using DFS-R to replicate the volume to a mirror server as
an extra layer of redundancy.
|
|
You loose the equivalent of one Physical Drive per each RAID5 Logical Drive
you create. So if you create two you loose the space of 2 physical drives.
So use one RAID5 Logical Drive consisting of 14 active physical drive with
one hotspare.
Partition it into one large Volume if NTFS will go that high
(13-Tb),...sorry I don't remember the limit or if there is a limit. If it
won't go that high then go as high as you can and partition the rest as a
second Volume and use it for other unrelated things (like file-based Server
Backups maybe).
Rebuild time is irrelevant,...why?
1. You have a hot spare anyway,...so the rebuild could take a month
and it wouldn't matter.
2. Rebuild time is directly related to the physical size and
read/write
speed of the physical drive. How it is partitioned doesn't mean
squat.
--
Phillip Windell
www.wandtv.com
The views expressed, are my own and not those of my employer, or Microsoft,
or anyone else associated with me, including my cats.
-----------------------------------------------------
"Barkley Bees" <barkleybees[ at ]hotmail.com> wrote in message
news:eDMzMcxTJHA.4680[ at ]TK2MSFTNGP06.phx.gbl...
[Quoted Text] > I'm going to be setting up a PowerVault MD3000 enclsoure with an NF500
> PowerEdge server. The enclosure will have 15 x 1TB SATAII drives
> (7200rpm). I plan to use one drive as a hotspare but my question/concern
> is regarding the RAID configuration.
>
> Should I:
>
> 1. split the configuration into two separate RAID5 packs (7 drives -
> RAID5, 7 drives - RAID5, 1 drive - global hotspare) and thus two volumes.
>
> 2. keep it as one large volume (14 drives - RAID5, 1 drive - hotspare).
>
> My preference is option #2 as I will be using this storage purely for user
> home directories so it would be much simpler from a management point of
> view (configuration of home path in Active Directory -
> \\servername\share\username). But my concern with option #2 is the rebuild
> time in the instance of a drive failure.
>
> Has anyone had any experience in this area that can offer an opinion?
> Thanks.
>
> *note: will be using GPT to span the 2 TB volumes.
> *note 2: will be using DFS-R to replicate the volume to a mirror server as
> an extra layer of redundancy.
>
>
>
|
|
Actually, I wouldn't do that.
During a rebuild there is tremendous stress on your drives. With only
basic RAID5 and a large stripe of spindles you increase the odds of
losing a second drive. With RAID 5 if you lose a second drive while the
rebuild is in progress, you will lose your data. Remember there is
only one parity block written per stripe. If your rebuild takes a
month, cross your fingers that you don't lose a second drive during that
month.
If you have the ability use a more redundant system such as two RAID5
arrays or RAID 10 (uses 50% disk space) or RAID6, then do that.
~kenny
e one RAID5 Logical Drive consisting of 14 active physical drive with
[Quoted Text] > one hotspare.
> Partition it into one large Volume if NTFS will go that high
> (13-Tb),...sorry I don't remember the limit or if there is a limit. If it
> won't go that high then go as high as you can and partition the rest as a
> second Volume and use it for other unrelated things (like file-based Server
> Backups maybe).
>
> Rebuild time is irrelevant,...why?
> 1. You have a hot spare anyway,...so the rebuild could take a month
> and it wouldn't matter.
> 2. Rebuild time is directly related to the physical size and
> read/write
> speed of the physical drive. How it is partitioned doesn't mean
> squat.
>
>
|
|
"Kenny Speer" <kenny.speer[ at ]gmail.com> wrote in message
news:ORvhUeMVJHA.3912[ at ]TK2MSFTNGP06.phx.gbl...
[Quoted Text] > Actually, I wouldn't do that.
>
> During a rebuild there is tremendous stress on your drives. With only
> basic RAID5 and a large stripe of spindles you increase the odds of losing
> a second drive. With RAID 5 if you lose a second drive while the rebuild
> is in progress, you will lose your data. Remember there is only one
> parity block written per stripe. If your rebuild takes a month, cross
> your fingers that you don't lose a second drive during that month.
Well I didn't mean a "month" that litterally :-)
Yes, I see what you are saying. Well on my servers I use RAID5 like I
described and the rebuild isn't that bad. It's actualy RAID1 on the OS Drive
and RAID5 on the Data Drive. If I lose another drive during rebuild and lose
the array,..well, that is what I have the daily and weekly backups for.
But one of our big arrays on our Video Server that works with our video
automation (we are a TVstation) is two identical fiber-channel External
Arrays with duel power supplies in each. Each unit is a RAID5 with a hot
spare,...but then the two units as a whole are Mirrored against each other
(like RAID1) so we would have to loose several drives on both units to loose
the whole content. The chances are greater that an operator will foul up the
content directly than the chances of losing it to hardware failure.
The content has a short life-span for the most part and so there are no
additional backups made of it. Besides it would take so long to
backup/restore that much data that it wouldn't be practical. If we lost it
we would just start dubbing tapes back into it from the point that we are at
and run the station off of tapes until the Server had enough content to
start playing on its own,..and then dub the rest while it is running.
--
Phillip Windell
www.wandtv.com
The views expressed, are my own and not those of my employer, or Microsoft,
or anyone else associated with me, including my cats.
-----------------------------------------------------
|
|
There are three areas to take into consideration
1. The MD3000's "Disk Groups"
2. The partitions you make using Microsoft's "Disk management"
3. Your amount of time backups and restores can take.
1. If you are going to go with option 2, you should try to use RAID 6 which
is a new option for the MD3000 that became available 12/19/2008. RAID 6 can
tolerate the loss of two disks without loosing any of your data. This is
important because with very large Disk Groups it could take weeks to rebuild
to your hot spare after a drive fails. If you lose a second disk during that
time your data is history.
2. Have you ever tried to run chkdsk on a 1.6TB partition with 5 million
files after you notice some corruption? I have, and it took over 3 days.
Since running Chkdsk /f doesn't allow access to the data while it runs, I was
out of commission for 3 whole days. I learned after that it's important to
take into consideration how long a chkdsk will take if you have corruption in
the filesystem. I'd say 2 million files is the most a single partition
should hold if you want to make sure chkdsk's don't take more than a day.
Chkdsk is affected by the speed of your array, mine was a RAID 5 with 14 10k
RPM disks. If you use 15k RPM or especially if you use SSD, then you could
have many millions more files in a partition. Too bad SSD isn't really
affordable enough for file server use yet.
3. If you have two separate Disk Groups you should be able to backup both
at the same time, meaning you can backup almost twice as fast. If you just
allocate two backup agents to the same Disk Group of physical disks it will
take the same amount of time or will actually be slower because of the
contention caused by having both agents asking for different data. The HDD
heads have to seek all over the place. For restore it's the same thing, your
restore speeds will be limited by disk contention.
In general I'd go for option 1. Being cheap always costs more later on.
I'd even do option 1 with RAID 6 instead of 5. You're the only one who
really knows whats going on, and being penny wise and pound foolish will only
make you look bad when something goes wrong later. Ask for what is ideal,
let someone else force you to compromise. That way you did your best.
Ganbatte!
"Barkley Bees" wrote:
[Quoted Text] > I'm going to be setting up a PowerVault MD3000 enclsoure with an NF500
> PowerEdge server. The enclosure will have 15 x 1TB SATAII drives (7200rpm).
> I plan to use one drive as a hotspare but my question/concern is regarding
> the RAID configuration.
>
> Should I:
>
> 1. split the configuration into two separate RAID5 packs (7 drives - RAID5,
> 7 drives - RAID5, 1 drive - global hotspare) and thus two volumes.
>
> 2. keep it as one large volume (14 drives - RAID5, 1 drive - hotspare).
>
> My preference is option #2 as I will be using this storage purely for user
> home directories so it would be much simpler from a management point of view
> (configuration of home path in Active Directory -
> \\servername\share\username). But my concern with option #2 is the rebuild
> time in the instance of a drive failure.
>
> Has anyone had any experience in this area that can offer an opinion?
> Thanks.
>
> *note: will be using GPT to span the 2 TB volumes.
> *note 2: will be using DFS-R to replicate the volume to a mirror server as
> an extra layer of redundancy.
>
>
>
>
|
|
|