[Date Prev][Date Next] [Thread Prev][Thread Next]
Re: [K12OSN] Raid Level Choice
- From: Liam Marshall <lsrpm mts net>
- To: "Support list for opensource software in schools." <k12osn redhat com>
- Subject: Re: [K12OSN] Raid Level Choice
- Date: Thu, 23 Dec 2004 14:27:53 -0600
On Wed, 2004-12-22 at 16:58 -0600, Les Mikesell wrote:
> On Wed, 2004-12-22 at 16:04, Liam Marshall wrote:
> > so are you saying there is basically no advantage to hardware raid?
> Not quite. If you use RAID5 it is better done in hardware, but you
> need one of the high-end controllers (3ware etc.) that really does
> the work in hardware instead of mostly in the drivers. RAID1 does
> not involve much computation so there is not a big advantage to
> dedicated hardware over just having each drive on it's own controller
> (always true in SATA) so the access runs in parallel.
> > I have effectively shot my wad getting the 4 drives and the raid
> > controller. I was intending it to be both increasing capacity and
> > giving the system some sense of disaster recovery through raids ability
> > to rebuid or work with alternate mirrored drives
> You will accomplish that any way you approach it. I tend to not
> trust new hardware until I've tested it myself and I especially
> don't trust new hardware with Linux drivers that don't have a
> lot of real-world testing. I have done software RAID1 on an
> assortment of drive types and controllers and have replaced
> failed drives with the 'mdadm' tool, so I know that works. I
> also have some 3ware controllers that so far have not had any
> problems. With anything else I'd either have to do some extensive
> testing of how the drive failure/rebuild works or find where someone
> else had documented that kind of testing.
> > The setup I am currently using has less than 30 GB of hard drive space,
> > all of it slow, 5400 rpm old scsi drives. The faster newer larger SATA
> > drives should give me at least the equal to it performance wise. I can
> > function now, I am just out of space and suspecting a drive of being on
> > the verge of failure. each new drive is 80 GB in size, so even in
> > something like raid 10 I am getting 160 GB usable space, right? over 5
> > times what I have now. with raid 5 I would have even more.
> You'll lose about a third of your space to raid5 compared to half for
> > What I am looking to do is acheive a perfomance increase, however
> > slight, while insuring a better disaster recovery.
> You should get that whether you run the controller in JBOD
> (Just-a-Bunch-Of-Disks) mode with software raid or let it
> do RAID5.
> > the controller I bought does hardware raid, so why wouldn;t I use that
> > instead of software raid via lvm, which in itself has a performance hit,
> > right?
> First you have to be sure that the controller is going to do all
> the work. Some of the SATA RAID controllers really do the work in
> the drivers and since the drivers are new and less tested than Linux
> MD arrays they are likely to be less efficient and more prone to bugs.
> Also, look at what you have to do to rebuild the raid after a drive
> fails - with some you would have to shut down for a fairly long
> time to do it in the controller firmware.
> > I just don;t know for sure whether to do 5 or 10. People tell
> > me 5 is better, disaster recovery wise, but with a performance hit. 10
> > is faster, but no parity stuff is happening, so less disaster recovery
> That's not quite true. Raid5 uses a parity disk and computation to
> fill in missing data. Raid 1 keeps 2 full copies. You can only
> lose one disk at a time out of a raid5 set and access to the set
> is likely to be slow until it is rebuilt. With raid1 you can lose
> more than one drive as long as each one that fails still has a
> working mate, and there is no speed penalty to having failed drives
> in the set.
> If your controller really does do the work in hardware and offers
> to automatically rebuild from a hot spare, another option besides
> what you've mentioned would be to put 3 drives in a raid5 set with
> another reserved as a hot spare. This would give you less space
> and less performance but would take care of itself completely when
> a disk fails. The opposite approach, if you don't trust the controller
> and want to keep things simple is to make 2 RAID1 sets, mounting one
> of them as /home. Unless you need more space than a single drive
> holds in /home you wouldn't need the extra complexity of raid0 striping
> or LVM - just put /boot and / one one device and /home on the other.
> Or are the SCSI drives going to stay in the machine too?
> Les Mikesell
> les futuresource com
just finished the basics now.
I went with using hardware raid
raid 5 utilizing 3 of the 80 GB drives and designating 1 80 GB drive as
K12LTSP 4.2.0 rc1 recognized raid controller with nothing for me to do.
Installed really fast too
after installing Staroffice and a few sundries, I booted the
workstations. All 32 connecting!
Of course, I can't test the load 32 kids put on it, but it seems faster
and as a bonus, my usb portable hard drive (40 GB maxtor drive in
enclosure) works on the server hotplugging
seems things are roses here
after testing workstation functionality I followed a how to on restoring
users and their home directories and mail to the new setup. This also
worked! Something has to go wrong. This is just not normal! :)
[Date Prev][Date Next] [Thread Prev][Thread Next]