Re: 16 port useful SATA controller
From: Abdullah Ramazanoglu (abdullah_at_ramazanoglu.tr)
Date: 12/15/04
- Next message: General Schvantzkoph: "Re: wireless-ready dsl gateway"
- Previous message: Brian Hall: "Re: Here's how to enable 1T Command Rate for K8T Neo FIS2R MS-6702 in Linux"
- In reply to: Rex Dieter: "Re: 16 port useful SATA controller"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Wed, 15 Dec 2004 21:41:01 +0200
begin Rex Dieter dedi ki:
> kermit wrote:
>> Rex Dieter wrote:
>>
>>>Nothing prevents a pure software RAID-5
Agreed. And that's what I would suggest to the OP.
>> except performance issues. I have never seen anybody using it in real life.
>
> Performance-wise, with a beefy CPU, software raid will oftentimes trump
> hardware raid. IMO, it's the failover/possible-hotswap support that
> makes hardware raid appealling.
There are next to no performance issues with software raid-5. I'm running
a 10 disk (9 raid + 1 hot-spare) 1 TB raid5 array with negligible CPU
overhead. See my relevant message:
Msg-Id: <2unpu1F2brrqgU1@uni-berlin.de> or,
http://groups.google.com/groups?as_umsgid=2unpu1F2brrqgU1@uni-berlin.de
BTW, the <10% CPU utilization is due to one of the disks running in
DMA disabled mode, which by itself accounts for more than 5%. This leaves
us with <5% CPU overhead on an Athlon 2400+ for a 10 disk 1TB raid-5 array.
As for the hot-spare and hot-swap issues, the former one is already being
handled by software raid on Linux. The latter is also handled, provided
that you have hardware-wise arrangements for hot-swap. E.g. if you have
SATA disks (which I believe support hotswapping hardware-wise). To hotswap
a disk you would do:
1. raidsetfaulty /dev/md0 /dev/hdc1 [*]
2. raidhotremove /dev/md0 /dev/hdc1
3. Physically detach the old disk and attach the new one (with same
geometry and partition structure) in its place.
4. raidhotadd /dev/md0 /dev/hdc1
[*] If it's not already set as faulty by md, i.e. if it is a healthy and
running disk. In this case, setting /dev/hdc1 as faulty will instantly
kick-in the hot-spare, and a resync process will automatically start to
sync the hot-spare in. This may take a while, and since you would want to
resync the hdc back after you've swapped it, you wouldn't want the
hot-spare to kick in. To prevent this behavior you should have already set
the hot spare as faulty and removed it from the array, with:
0a. raidsetfaulty /dev/md0 /dev/hdq1 (assuming hdq is the hot-spare disk)
0b. raidhotremove /dev/md0 /dev/hdq1
and after resyncing of new hdc is started at step 4 (no need to wait for
resync to complete), you can add the hot-spare back into the array:
5. raidhotadd /dev/md0 /dev/hdq1
-- Abdullah | aramazan@ | Ramazanoglu | myrealbox | ________________| D.0.T cöm |__
- Next message: General Schvantzkoph: "Re: wireless-ready dsl gateway"
- Previous message: Brian Hall: "Re: Here's how to enable 1T Command Rate for K8T Neo FIS2R MS-6702 in Linux"
- In reply to: Rex Dieter: "Re: 16 port useful SATA controller"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
