RAID5 Sucking? 4 x 500GB 7200RPM Drives

[Roger H. Veteran]

Ok so i got this Foxconn Board (Digital Life x38A) with the Intel ICH9R soutbridge for RAID. I got 4 Seagate drives at 500GB 7200RPM 32MB Cache and this thing is only pushing like 70MB/s. I did tests with HDTune and HD Tach and thta's all ti says.

I tried XP 32bit, Vista 32bit and Vista 64Bit and they all give me crappy speeds. Oddly enough tho, Vistax64 gives 5.9 scores but 32bit gives only 4.8.

The system includes.

800 Watt

2 x 8800GTX

Foxconn Board mentioned above

8GB 800Mhz DDR2

4 x 500GB 7200RPM 32MB Cache

2 x DVDRW Drives

[+PeterUK MVC]

Are the SATA/150 jumper off? Drivers updated?

[Roger H. Veteran]

I tried RAID10 and I removed the jumpers and yeah it jumped up to 207MB/s MAX in HDTUNE which was awesome then i updated drivers and but it sucked like crazy.

I'm gonna go test some other stuff to see what else could be the cause. What am i supposed to be getting anyways? I saw some tests and their burst speeds were like 3,300MB/s or something.

That plus the chart is all over the place... it goes up then down and all that. Not a steady stream or somehting like that.

[jamend]

RAID5 is not that fast... I get about the same on my 4x250GB.

[Roger H. Veteran]

http://techreport.com/forums/viewtopic.php?p=769864#p769864

Well that guy seems to be getting some good speeds but that's with HDTach, i don't anything close to that tho.

[icie]

Generally, I found RAID 5 to be sucky too. Not sure why, but might be due to the lower random write speeds affecting operations. If anyone find there's a definitive problem with the configuration (and a solution), I'd be interested.

[neufuse Veteran]

RAID5 has a LOT of overhead on it... all the parity writes the multi drive spanning writes all that fun stuff... RAID5 don't expect performance, it's there for data integirty... thats why when you set up stuff like SQL Servers that have high performance requirements you set it up in RAID1+0 not RAID5 because RAID5 is slow, RAID6 is even slower yet (RAID6 IS RAID5 + ADG (basically dual parity))

[shuntsu]

Generally, I found RAID 5 to be sucky too. Not sure why, but might be due to the lower random write speeds affecting operations. If anyone find there's a definitive problem with the configuration (and a solution), I'd be interested.

Well, basically RAID5 isn't that fast, period, no matter how you build it. RAID5 requires parity information to be written for every write operation performed, and it just kills performance. The parity striping also increases in cost the more disks are in the array. You still get pretty good parallel read performance, but that's it.

RAID10 is much faster, because there is no parity information. Each stripe is simply written twice (in parallel), so you get basically (number of disks) / 2 worth of write performance, which typically blazes past RAID5. Read performance is similar.

[neufuse Veteran]

http://techreport.com/forums/viewtopic.php?p=769864#p769864

Well that guy seems to be getting some good speeds but that's with HDTach, i don't anything close to that tho.

Something about those graphs doesn't seem right... at work we have our file server on high end RAID controllers with 7 drives per array in RAID5 and we don't get speeds like that... (7 15K RPM HDD's at 146.8GB each connected to a dedicated RAID5 controller with 512MB Battery backed up Cache connected with dual 3Gbps fiber channels) and if we can't get those speeds no idea how they did... we get about 80MB/ps with 100MB/ps burst rates and about 12ms access times the small file IO (say 4KB blocks) we get horrible performance about 2MB/ps write and 8MB/ps read.....

[jamend]

Something about those graphs doesn't seem right... at work we have our file server on high end RAID controllers with 7 drives per array in RAID5 and we don't get speeds like that... (7 15K RPM HDD's at 146.8GB each connected to a dedicated RAID5 controller with 512MB Battery backed up Cache connected with dual 3Gbps fiber channels) and if we can't get those speeds no idea how they did... we get about 80MB/ps with 100MB/ps burst rates and about 12ms access times the small file IO (say 4KB blocks) we get horrible performance about 2MB/ps write and 8MB/ps read.....

If "2MB/ps" means 2 MB/s, then you must be doing something wrong...

KoDeXeRo, what stripe size did you use?

[+PeterUK MVC]

RAID5 is not that fast... I get about the same on my 4x250GB.

RAID 5 with three disks easily gets 100MB/s + so its not slow.

I get an average of 120MB/s & 131MB/s read with over 2000MB/s burst speed so there's no reason why someone with four disks & a newer Intel ICH controller to not get better results.

[neufuse Veteran]

If "2MB/ps" means 2 MB/s, then you must be doing something wrong...

KoDeXeRo, what stripe size did you use?

Have you ever benchmarked a RAID5 drive on all different types of read/write levels? because that is exactly what it should be for that size... RAID5 pretty much sucks for small block sizes in read/write... say you have a 16KB stripe size with RAID5 over 7 disks... you get these speeds

SIZE

4KB (Random IO) - 2MB/s - 6MB/s

1MB (Sequential) - 30MB/s - 50MB/s

10MB - 45MB/s - 80MB/s

100MB - 50MB/s 88MB/s

that is Write speed / read speed

4KB blocks across a random read / write will always be slow, but its horribly slow in RIAD5... RAID5 drives need frequent defragmentation to keep them in good condition speed wise in a large file store environment like ours... the average file size in our file server is about 16-35KB and the largest stripe we can make is 64KB but we opted for RAID6 instead for increased security (you can lose 2 drives in RAID6 before data loss) and the largest stripe on RAID6 with a 400GB logical drive is 16KB on our controller... but there wasnt too much a difference we we benchmarked the RAID5 config... the speeds above were from RAID5...

RAID 5 with three disks easily gets 100MB/s + so its not slow.

I get an average of 120MB/s & 131MB/s read with over 2000MB/s burst speed so there's no reason why someone with four disks & a newer Intel ICH controller to not get better results.

yeah but how did you benchmark it? sequential IO, random IO, how large is you stripe size? how big are the allocation units on your drive when you formatted it? How large of a file did you benchmark against? small files perform horribly on RAID5 large files work relatively well

[+PeterUK MVC]

Think I found your problem with RAID 5 (something I did before I run benchmarks).

In Intel? Matrix Storage Manager switch to advanced mode right click on your Volume and click Enable-Volume write-back cache.

This should make it run fast.

[+PeterUK MVC]

yeah but how did you benchmark it? sequential IO, random IO, how large is you stripe size? how big are the allocation units on your drive when you formatted it? How large of a file did you benchmark against? small files perform horribly on RAID5 large files work relatively well

HD Tach run on the 32mb zones test stripe size is 64KB format using 4096 bytes allocation units.

I have not tested small file performance but it will not be fast which I did know.

[neufuse Veteran]

HD Tach run on the 32mb zones test stripe size is 64KB format using 4096 bytes allocation units.

I have not tested small file performance but it will not be fast which I did know.

That'd be why you see better speeds, you are using larger segments to benchmark... to get a true benchmark start with 4KB read/writes then work up to stripe size (what ever that is) then test larger ones... then average them out

[Roger H. Veteran]

If "2MB/ps" means 2 MB/s, then you must be doing something wrong...

KoDeXeRo, what stripe size did you use?

I was using 128K and then tried 64K stripes and it was about the same. I'm doing RAID10 now and just gonna keep that. I did the Enable Write-Back Cache thingy the guy mentioned above but right now i don't see much of a boost (yet)

I'm using the same now as PeterUK and it seems that's where i get 207MB max so far.

(Oh and this is under Vista 64bit since the machine is 8GB of Corsair CS4 RAM - didn't want to waste that on XP Pro - When i told him that XP 64bit sux and Vista is basically his only choice he was nervous but so far he says he's liking Vista)

[shakey_snake]

Yeah, you shouldn't try to run your OS from RAID 5.

If RAID 10 is working for you, sweet.

But personally, given your info in the first post, I'd set up 2 volumes. one would consist of 3 disk RAID 5 for secure storage and the other would be 1 disk for you OS and programs.

Granted, you OS and programs wouldn't be securely stored, bit you'd gain 500GB of space over RAID10.

[Roger H. Veteran]

well hehe. he has like 3 External 500GB Mybooks so data gets dumped there once it's done so he'll just put them there. He's running his stuff there to test it out and see if it's good enough.

[+PeterUK MVC]

That'd be why you see better speeds, you are using larger segments to benchmark... to get a true benchmark start with 4KB read/writes then work up to stripe size (what ever that is) then test larger ones... then average them out

Ok would help to know what benchmark you used but here is a compare of RAID 5 vs WD5000AAKS using CrystalDiskMark.

Yes write speed is slow but for loading/reading it a bit faster and for home/gaming use thats whats needed, as for servers its not so good for lots of write access but your 4KB test is much better but that's over 7 disks.

Edited April 7, 2008 by PeterUK