RAID Software Experiments
- From: ivowel@xxxxxxxxx
- Date: 23 Feb 2006 08:31:01 -0800
I thought others may find this information to be interesting. I did
some lower-level tests than what the RAID howto reports, i.e., more
filesystem independent.
I started experimenting with linux software RAID. For the most part,
it works very nicely and without much trouble. I mostly followed of
the information in
http://www.tldp.org/HOWTO/Software-RAID-HOWTO-5.html#ss5.8. It is very
helpful, but it could use some updating (the RAID-1 raidtab file needs
a chunksize; ext3 is the more common choice over ext2; the link between
the mke2fs parameters and the raid parameters is not 100%
clear---actually, mkfs.ext3 should be smart enough to look at the
raidtab file if the device is a /dev/md* device, but it may not.) It
reads like a document written 5 years ago, with minor updating here and
there. (No, I am not volunteering to do it better. Just wanted to
note this; I am grateful that the doc was there to begin with.)
My system is an old Athlon64 3000, 1.5GB of RAM. It has two Raptors
34GB (sda and sdb) on an onboard sata_via w/ own irq, and one subraid
(hda), which is really an encapsulated transparent hardware RAID-1
device. It contains two SATA 300GB drives. The subraid is very nice,
but not smart enough to handle NCQ, nor capable of running at either
RAID-0 or RAID-1. It is purely RAID-1.
The experiments now rely solely on "hdparm -t" output. As such, they
are impervious to filesystem issues. I ran each a couple of times and
chose some representative samples. for the most part, perhaps except
the first run, the runs came up with similar performance.
----------------------------------------------------------------
Background
hda subraid w/ 2 300GB SATA drives (no linux software raid)
Timing buffered disk reads: 64 MB in 1.34 seconds = 47.84 MB/sec
Timing buffered disk reads: 64 MB in 1.25 seconds = 51.04 MB/sec
sda raptor 34GB
Timing buffered disk reads: 64 MB in 1.17 seconds = 54.57 MB/sec
Timing buffered disk reads: 64 MB in 1.17 seconds = 54.57 MB/sec
sdb raptor 34GB
Timing buffered disk reads: 64 MB in 1.21 seconds = 52.73 MB/sec
Timing buffered disk reads: 64 MB in 1.17 seconds = 54.57 MB/sec
--> All drives have roughly equal performance. Obviously, the hda is a
very different animal.
----------------------------------------------------------------
linux software raid experiments, literally copying the raid parameters
from the software raid howto:
md0 RAID0*raptor 34GB sda2,sdb2, each 8GB
Timing buffered disk reads: 64 MB in 0.84 seconds = 76.20 MB/sec
Timing buffered disk reads: 64 MB in 0.81 seconds = 79.12 MB/sec
md0 RAID1*raptor 34GB sda2,sdb2, each 8GB [chunksize 4]
Timing buffered disk reads: 64 MB in 1.35 seconds = 47.38 MB/sec
Timing buffered disk reads: 64 MB in 1.28 seconds = 49.81 MB/sec
md0 RAID5*raptor 34GB + subraid sda2,sdb2, each 8GB, plus hda3 32GB
as parity
Timing buffered disk reads: 64 MB in 6.29 seconds = 10.17 MB/sec
----------------
all cache reads (-T) on all drives are something like
Timing buffer-cache reads: 128 MB in 0.43 seconds =295.66 MB/sec
obviously, I would have preferred running this with three RAPTOR
drives, but I did not have one. the sda drives had equal sized
partitions on 2, both beginning at 6 and ending at 1002 (8GB in size,
each). the RAID5 experiment may have suffered from the fact that hda3
started at block 8193 (of 36481). All drives were very mildly engaged
otherwise. hda contains the running gentoo linux system drive and sda3
is the home directory. none of the drives were very busy, but they
were not completely idle either.
NOTES: before filesystem issues,
running drives at RAID0 increases performance by about 50% (nothing
close to 100%). surprising---the two sata channels should have been
independent and the CPU should not have been constraining.
running drives in RAID1 degrades the performance by about 10%. a mild
surprise.
on my system, RAID5 in a mixed setup has unacceptably poor performance.
Hopes this helps others...curious whether this is a representative
experience (especially the RAID-5).
sincerely,
/iaw
.
- Follow-Ups:
- Re: RAID Software Experiments
- From: John-Paul Stewart
- Re: RAID Software Experiments
- From: Grant
- Re: RAID Software Experiments
- Prev by Date: Urgent Pls! Unable to display thru "Level One" brand "FCS-1010" model IP webcam
- Next by Date: Re: RAID Software Experiments
- Previous by thread: Urgent Pls! Unable to display thru "Level One" brand "FCS-1010" model IP webcam
- Next by thread: Re: RAID Software Experiments
- Index(es):
Relevant Pages
|
