Sign in with
Sign up | Sign in
Your question

poor performance on RAID5

Tags:
  • Performance
  • Hitachi
  • NAS / RAID
  • Storage
Last response: in Storage
Share
February 7, 2007 10:01:36 AM

I have a problem with my 3x160GB hitachi RAID 5- Gigabyte DS4 ICH8R - poor READ performance, I didn't try IOMETER yet, but I wonder why here
http://forumz.tomshardware.com/hardware/PCI-Express-SAT...
are the HDtach results correct on the same chipset?
Do You have any suggestion how to measure it via HDtach? And why is the read speed so low?

could be problematic that my 160GB Seagate is on channel 0, hitachi on 1,2,3??

or because of the array wasn't initialized yet? (some guy wrote to me, after two days or so the array starts to count parity...)

More about : poor performance raid5

February 7, 2007 3:21:33 PM

the poor performance is because you are using the bare minimun setup for raid5. in the test you link to the system contained 4 drives. Raid5 increases performance with each drive you add to the array. Your raid5 with 3 disks should be slower than the test you linked to.

Raid5 takes a little while to build parity once it is initially setup. The guy gave you a very conservative time frame of 2 days, which sounds ideal for a large array.
February 7, 2007 4:29:48 PM

ok, and how do you explain the difference in burst read, and diffence in average read with test below? it's test from this afternoon, the test before has been made 24 hours ago... but still Average Read is smaller than I would expect.

and the guy wrote that two days AFTER he set up RAID, ich8r STARTED to migrate (count parity) - he didn't say it took 2 days. Don't know why INTEL do it this way, perhaps they think first day is HDD too busy because of installation of aplications?
he has average read 191MBps with 4 disks (7200rpm WD). Is it possible one disk could make such a difference (90MBps)?

Related resources
February 7, 2007 8:39:54 PM

the performance increase for the second benchmark you showed could be explained by two factors.

1. The fourth drive will give the second benchmarks somewhere between 40-60 mb/s advantage over your setup.

2. Western Digital drives versus your Hitachi drives. I would need model numbers to say for certain but its more than likely the WD's have a speed advantage over the Hitachi line that you are using.
February 7, 2007 11:35:08 PM

you also did not mention what controller you are using(i think you might of done but its not clear)

the controller can make a large difference, heres my RAID 5 array results using 4 320GB seagate 7200.10 on a onboard controller(Silicon)

as you can see my burst speed is very low(not sure why ill try and fix that at some point) but my average read is 97.4MBs, i dont really expect alot more than that since i am using a onboard controller which is partionally software controled.



P.S

the other drive in the picture is my main drive that contains my OS, games and programs, the RAID has DVDs, TV Shows and Music and has around 20GB of free space left(not sure if space makes a difference might be the cause of the burst speed being low)


EDIT

a little tip, press ALT and printscreen to take a pic of just the window, just means you dont have to crop the pic to show just the result....much quicker.


EDIT #2

running the test while the raid array is still building will slow down performance i know ive tried when i first ran the test my average speed was only a mere 60MBs because the drives hadnt finished.

the difference in burst speed i believe corresponds to the size of the drives being used, as for the average read that can be affected by background programs, and the controller being used. Another tip for increasing performance, my drives came with a jumper that set them to work in sata 150 instead of 300, now i took that jumper out of all of them and it gave my average read a nice boost, even though the controller itself is only sata150, i can only summerize that raid 5 can take advantage of almost the full 150 speed.

sorry for all the edits i am very tired, and am now going to bed, theres probs a few spelling mistakes too, ill fix them in the morn. nyhow i hope you feel better bout your results now, and jusdt be happy you have a RAID 5 that gives better performance than a single drive(it could always be worse ukno)
February 8, 2007 5:46:41 AM

Quote:
nyhow i hope you feel better bout your results now, and jusdt be happy you have a RAID 5 that gives better performance than a single drive(it could always be worse ukno)


thats fine it's quicker, but I thought it should be a bit quicker. I'm a bit disapointed, but it seems it's top speed I can make with three drives...

It's onboard chipset - ICH8R (mentioned in the first message). I'm not aware HITACHI has the switch you mentioned, but It's strange that in the Intel Storage Manager is disks are running in SATAI mode, but disks are SATAII.

Quote:
the difference in burst speed i believe corresponds to the size of the drives being used, as for the average read that can be affected by background programs

True... tested on rebuilding array after disabling one of disks by cuting power to HDD. It's shame Intel didn't make Manager more user friendly - User Interface should keep user informed, what's running on the backgroud more thoroughly...

that guy don't have WD, he has 4x Seagate 7200.10 - 320BG, 16MB, SATA, - perpendicular recording - true - It's a bit better disk when I compare it with mine.
Thanks fot the tips.
Your results are awfull when I compare them with RAID5 with four disks the guy has-probably controller issue:

February 8, 2007 10:18:34 AM

Sigh, I see so many threads of people getting poor performance and then I see they're using onboard RAID. Onboard RAID is meant to be cheap, not fast or effective. Because no XOR engine is involved it has to use your CPU so it's slowing the computer down as a whole as well, plus the added latency of trying to make XOR calculations.

He has 4 PMR drives and you have 3 standard drives, that's a huge difference.
February 8, 2007 11:57:00 AM

Quote:
Sigh, I see so many threads of people getting poor performance and then I see they're using onboard RAID. Onboard RAID is meant to be cheap, not fast or effective. Because no XOR engine is involved it has to use your CPU so it's slowing the computer down as a whole as well, plus the added latency of trying to make XOR calculations.

He has 4 PMR drives and you have 3 standard drives, that's a huge difference.


well, and what would be the average read with non-onboard RAID with same disks? I didn't see tests...
February 12, 2007 1:54:33 PM

Why are you using raid5? Is it because you want the best redundancy? If you controller allows it and you can add a 4th drive then I would suggest using 1 +0. This will mirror drives 1 + 2 as the first pair and then 3 + 4 as the second pair, it will then mirror the two pairs of mirrored drives! (hope that makes sense).

The preformance is much better than with raid 5 and offers great redundancy as well. I have just moved alot of our servers to raid 1+0 and performance has improved.
February 12, 2007 7:11:37 PM

Isn't speed of 1+0 similar to speed of RAID5 with four disks?
February 12, 2007 9:00:38 PM

raid5 is only a little bit slower than Raid 0, so raid 1+0 would be a waste of time.

the writes in a raid 5 take slightly longer yet the read rate is about the same as Raid 0.
February 12, 2007 9:40:47 PM

mostly they are bells and whistles! Somewhat functional but not compared to the big guns!
February 13, 2007 7:33:51 AM

cool, everybody says "Somewhat functional but not compared to the big guns!", but nobody's able to prove it (tests with comparable disks). I love this guys:( 
February 13, 2007 1:28:03 PM

Quote:
cool, everybody says "Somewhat functional but not compared to the big guns!", but nobody's able to prove it (tests with comparable disks). I love this guys:( 






Here is a review that is basically what everyone is saying. I agree with nmarchini, when it comes down to it, it just depends on what you are after.

http://www.gamepc.com/labs/view_content.asp?id=r52005&p...
February 13, 2007 2:27:42 PM

OP, your second graph, for what it is, looks good to me. Is there a problem here?
February 13, 2007 3:13:27 PM

Quote:
cool, everybody says "Somewhat functional but not compared to the big guns!", but nobody's able to prove it (tests with comparable disks). I love this guys:( 






Here is a review that is basically what everyone is saying. I agree with nmarchini, when it comes down to it, it just depends on what you are after.

http://www.gamepc.com/labs/view_content.asp?id=r52005&p...

btw that review says the exact opposite, in all the tests in that review the ONBOARD silicon controller comes out top.....its the same controller im using lol.
February 13, 2007 4:28:34 PM

Quote:
cool, everybody says "Somewhat functional but not compared to the big guns!", but nobody's able to prove it (tests with comparable disks). I love this guys:( 






Here is a review that is basically what everyone is saying. I agree with nmarchini, when it comes down to it, it just depends on what you are after.

http://www.gamepc.com/labs/view_content.asp?id=r52005&p...

btw that review says the exact opposite, in all the tests in that review the ONBOARD silicon controller comes out top.....its the same controller im using lol.

Open mouth insert foot! I was looking for some test setups for supec and got ahead of myself! My mistake!
February 14, 2007 4:31:58 AM

Quote:
btw that review says the exact opposite, in all the tests in that review the ONBOARD silicon controller comes out top.....its the same controller im using lol.


No, double-check what the review says in the conclusion about write performance -- which is usually the key factor in picking between RAID 5 implementations. Any decent implementation can do RAID 5 reading well (or at least passably; don't trust HDTach on this), but some have very poor write performance.

Some on-board RAID 5 controllers have good write performance.
February 14, 2007 5:33:02 AM

Quote:
the difference in burst speed i believe corresponds to the size of the drives being used, as for the average read that can be affected by background programs, and the controller being used. Another tip for increasing performance, my drives came with a jumper that set them to work in sata 150 instead of 300, now i took that jumper out of all of them and it gave my average read a nice boost, even though the controller itself is only sata150, i can only summerize that raid 5 can take advantage of almost the full 150 speed.


Burst speed is a useless number, it doesn't have any real world bearing. Also, it doesn't matter wether it's SATA 150 or SATA 300 as each port has a dedicated 150MB/s or 300MB/s, the whole array doesn't share it. SATA 300 is just a gimmick.

To the OP, your reads look fine, as Madwand said pretty much anything can read RAID 5 fine but if you want good writes and overall performance you need to get a hardware controller, period.

Quote:
raid5 is only a little bit slower than Raid 0, so raid 1+0 would be a waste of time.

the writes in a raid 5 take slightly longer yet the read rate is about the same as Raid 0.


The reason RAID 1+0 is a good idea is because it doesn't involve parity and will give much better writes.

Quote:
Isn't speed of 1+0 similar to speed of RAID5 with four disks?


If you mean both setups have 4 disks, the RAID5 will have better sustained reads and the 1+0 will have better writes and latencies.

To sum things up, your results are perfectly fine for onboard RAID, if you aren't satisfied then get a real RAID controller. If you really wanna see benchmarks so bad tell me the conditions you want and I'll benchmark it on my hardware and onboard to prove it.
February 14, 2007 6:08:40 AM

Quote:
btw that review says the exact opposite, in all the tests in that review the ONBOARD silicon controller comes out top.....its the same controller im using lol.


No, double-check what the review says in the conclusion about write performance -- which is usually the key factor in picking between RAID 5 implementations. Any decent implementation can do RAID 5 reading well (or at least passably; don't trust HDTach on this), but some have very poor write performance.

Some on-board RAID 5 controllers have good write performance.

I think ICH8R isn't so bad in Write, at least according to this:
http://www.tomshardware.com/2007/01/03/the_southbridge_...
February 14, 2007 6:15:45 AM

Quote:
To sum things up, your results are perfectly fine for onboard RAID, if you aren't satisfied then get a real RAID controller. If you really wanna see benchmarks so bad tell me the conditions you want and I'll benchmark it on my hardware and onboard to prove it.


If it wouldn't be too much work for You... but when I imagine reinstalling system twice because of discussion only, I'd be too lazy to do it:) 
February 23, 2007 9:00:45 AM

I ran across this thread looking to see how my newly built RAID 5 NAS was actually performing. I am kind of disappointed by my results after looking at the original posters. My NAS consists of:

Intel Micro-ATX MB
Pentium4 2.4Ghz CPU
1GB (512 x 2) Samsung RAM
Seagate 20Gb HDD (for OS)
4x SATA II Seagate 400Gb drives (16Mb cache, perp. recording)
Highpoint Rocketraid 1740 4 port SATA II PCI Card



Is this what you would expect from a hardware RAID controller? The results are quite similar to the original poster using the integrated software RAID solution.
February 23, 2007 12:56:51 PM

Quote:
Is this what you would expect from a hardware RAID controller?


This is a great example of performance you would expect from a card that's bottlenecked on the PCI bus. Perhaps it could be a bit higher -- say 110 MB/s instead of 102 MB/s, but that's a minor detail. The graph shows a bandwidth limitation; the standard PCI bus has a similar bandwidth limitation.
February 23, 2007 10:45:34 PM

I was hoping that this was not the PCI bus. Oh well. I had to use it because this old mobo I am using doesnt have PCI-X or PCIE slots. The parts I did have to buy cost around $800 as it was so a new board was out of the question. Now I know where the bottleneck is. Thanks for your input.
February 26, 2007 4:58:51 AM

On second thought, Does anyone think the bottleneck might be the integrated ethernet adapter on the mobo? Just thinking because "theoretical" bandwidth should be about 133MB/s. Correct?
I could just buy a PCI Gigabit card and see if I can increase the throughput. Do you think this is possible?
February 26, 2007 1:51:28 PM

The benchmark you ran did not have anything to do with your ethernet card.

The PCI bus is often times a shared resource and there are other things in your system that utilize the PCI bus. So it is unlikely that you can get the full 133mb/s out that one PCI slot.

If you added a gigabit ethernet card then the only place to install it would be the PCI slot and that would in turn eat up more of the PCI bus.
!