brien dieterle wrote:
> 
>> I need your opinion on Linux software RAID. I mean its performace (in
>> terms of drive read/write time). I've configured Software RAID on two
>> SATA drives. I figured out the onboard SATA raid controller that this
>> server has is a fakeraid controller (Works only with the manufacturer
>> drivers on M$ Windoze). So, resorted to Linux software RAID. I've had
>> bad experience in the past with software RAID with 2.4 kernel based
>> Linux server. So, I'd like to know if anyone here has experiences with
>> software RAID on 2.6 kernel (default FC3 kernel).
>>
>> And by accident, I created three RAID devices (namely md0, md1 and md2
>> for /boot, / and swap respectively). Do I need to create RAID for swap
>> parition. Now, I'm worried that it's going to affect the performance. As
>> I write, it's still going thru FC3 installation process. Probably, I'd
>> perform some harddrive read/write access tests after installation. I'd
>> really appreciate if you can share your opinions and experiences with
>> software RAID in Linux 2.6 kernel (to be specific 2.6.9-smp).
>>
>> Thanks,
>> Sanjay.
>>
> 
> I've been running software raid5 on / (with 3 ide drives, not SATA), for 
> many months now with no problems whatsoever.  In fact, I moved the array 
> to a new machine and rearranged the drives onto different controllers 
> without a hiccup.
> 
> Swap on raid is probably recommended (if the swap device croaks while 
> the system is using it bad things can happen).  But, you are more than 
> welcome to add the partitions in fstab and the kernel will stripe to all 
> of them automatically, so no, raid is not required.
> 
> Your raid1 (you didn't make raid0, did you?) device will probably be 
> slightly slower than a single disk for reads, while a bit slower still 
> for writes.  I've done some benchmarking of several software and one 
> hardware raid systems, with less than amazing results.

I set up a RH9 server with software raid1 on ide a little over a year 
ago, and haven't had a problem (knock wood). The partition/definition 
during installation was a tad tricky, but straight forward. In order to 
provide fault tolerance, I decided to mirror the swap too.

I haven't done any performance measurements, but speeds aren't very 
noticably different than what I'd expect to see on a single drive. In 
theory, writes will be slower, and reads should be *faster*. I think 
it's curious that Brien said that raid1 reads would be slightly slower. 
I suppose that could be, but is not what I would expect. Meaningful 
comparisons haven't been possible for me to do, due to lack of 
equivalent hardware and time. ;)

> 
> Please forgive the HTML below-- here is a table of some benchmarked 
> systems.  The fastest machine was Hardware Raid, but it also had 4 disks 
> and much beefier hardware than most of the others.  It's hard to tell-- 
> I'm mostly interested in block read/write, so I ommited other results to 
> save space here.
> 
> 									
> 	
> 	Sequential Output (writing to disk) 	Sequential Input (reading from disk)
> 
> 	
> 	
> 	
> 	
> 	
> 	
> 	
> 	
> 	
> 	
> 	Per Chr 	block 	rewrite 	Per Chr 	Block
> host 	Kernel 	RaidType 	raidlevel 	#disks 	DiskType 	DiskSpeed (RPM) 
> CPUInfo 	mhz 	RAM(M) 	testsize 	K/sec 	K/sec 	K/sec 	K/sec 	K/sec
> nat1-single disk 	2.6.8 	none 	
> 	1 	ide 	5400 	(1) Pentium 165mhz 	165 	128 	6G 	1491 	10495 	5256 
> 1730 	11514
> Xena raid1 	2.4.22 	Hardware 	1 	2 	scsi 	10000 	(2) Intel(R) Pentium(R) 
> III 1400Mhz 	1400 	2000 	3G 	15682 	33073 	12604 	16394 	24242
> Fred-degraded 	2.6.7 	Software 	1 	1 	scsi 	10000 	(2) Pentium III 
> (Coppermine) 1.0Ghz 	1000 	2000 	10G 	14862 	20169 	11823 	16140 	28471
> Fred raid1 	2.6.7 	Software 	1 	2 	scsi 	10000 	(2) Pentium III 
> (Coppermine) 1.0Ghz 	1000 	2000 	10G 	14653 	19625 	11453 	16608 	28788
> Fred single disk 	2.6.7 	Software 	1 	1 	scsi 	10000 	(2) Pentium III 
> (Coppermine) 1.0Ghz 	1000 	2000 	10G 	14808 	22059 	12871 	16381 	31871
> Bud raid1 	2.4.22 	Software 	1 	2 	scsi 	10000 	(2) Intel(R) Xeon(TM) 
> CPU 3.06Ghz 	3000 	1000 	10G 	24288 	38629 	21496 	25724 	48377
> squirrel single disk 	2.6.9 	none 	
> 	1 	ide 	7200 	(1) Intel(R) Pentium(R) 4 CPU 3.00Ghz 	3000 	512 	10G 
> 33913 	35521 	18886 	26427 	52242
> brien-degraded 	2.6.9 	Software 	5 	2 	ide 	7200 	(2) Athlon XP 
> 1.53Ghz 	1500 	512 	10G 	18619 	20690 	16483 	26114 	52297
> brien-raid5 	2.6.9 	Software 	5 	3 	ide 	7200 	(2) Athlon XP 1.53Ghz 
> 1500 	512 	10G 	20756 	20477 	16935 	26764 	62454
> brien single disk 	2.6.9 	none 	
> 	1 	ide 	7200 	(2) Athlon XP 1.53Ghz 	1500 	512 	10G 	25492 	33957 
> 22044 	27278 	63052
> Gasp raid5 	2.4.22 	Hardware 	5 	4 	scsi 	10000 	(2) Intel(R) Xeon(TM) 
> CPU 2.80Ghz 	2800 	4000 	10G 	20936 	24329 	17461 	25829 	89098
> 
> 
> --brien
> 
>  


-- 
-Eric 'shubes'
"There is no such thing as the People;
  it is a collectivist myth.
  There are only individual citizens
  with individual wills
  and individual purposes."
-William E. Simon (1927-2000),
     Secretary of the Treasury (1974-1977)
  "A Time For Truth" (1978), pg. 237

 
---------------------------------------------------
PLUG-discuss mailing list - PLUG-discuss@lists.plug.phoenix.az.us
To subscribe, unsubscribe, or to change  you mail settings:
http://lists.PLUG.phoenix.az.us/mailman/listinfo/plug-discuss