Very poor RAID 5 performance, help!

PiKe

PiKe

PiKe

Caporegime
Joined
18 Oct 2002
Posts
25,287
Location
Lake District
Just build a RAID 5 server with 4 drives and 1 hot spare.

On doing the hdtach benchmark its giving almost 300MB/s burst but only 60/70MB/s sequential read, what's going on? It should be more like 240MB/s.

Can anyone help? :(
 
The onboard ICH8R.

hdtach.png


This is a benchmark from the same system except the hard drives are 10,000rpm and the CPU is quicker:

hdtach-ich8-raid5-big.gif
 
Onboard raid controllers are "soft-raid", basically all done in software/firmware. This means the CPU must do all the CRC calculations etc etc. RAID5 arrays are in any case never as fast as RAID0, so even with a fast dedicated (£££) RAID5 capable card you will never get close to the 240MB/sec you think you should (a RAID0 set might get close to that.) A faster CPU obviously will help, as will faster disks (which is probably why the Raptor set with faster CPU is so much faster, but it does seem disproportionate in my experience, especially for soft raid. Was that 4 Raptors in RAID5 or 3 in RAID0? The graph looks very much like 3 150's in RAID0 but I could very well be wrong.)
 
Last edited:
The raptors were in RAID5 aswell, but I'm sorry, 2500rpm isnt going to make the massive difference seen there, softraid or not.
 
In mine, C2D 1.80ghz, in the other one, Some E series chip with 4mb cache.

During testing the CPU utilisation in HDtach is only 3% as you can see.
 
Tried ATTO benchmark, transfer size 1024 length 8mb and got 635MB/s read and 25MB/s write?
 
Hmm, I just realised I should've been more careful - it's usually the *write* speeds that suffer in soft-raid setups due to the CRC calculations. Read only suffers if the array is in a degraded state (hence the data then has to be reconstructed on the fly.) So, since IIRC HDTach is a read-only benchmark I can't explain your poor read performance.
 
That was one avenue I explored but no, it wasn't that.

I went looking for the latest Intel drivers but that made no difference, then I poked around the intel storage manager and found that Volume-Write Back Cache wasn't enabled, enabled it and ran the test again and the results are what you can see above :)
 
PiKe said:
That was one avenue I explored but no, it wasn't that.

I went looking for the latest Intel drivers but that made no difference, then I poked around the intel storage manager and found that Volume-Write Back Cache wasn't enabled, enabled it and ran the test again and the results are what you can see above :)
Click to expand...
Holy crap, what a difference! Nice job :)

Oh, what make/model are the drives?
 
PiKe said:
That was one avenue I explored but no, it wasn't that.

I went looking for the latest Intel drivers but that made no difference, then I poked around the intel storage manager and found that Volume-Write Back Cache wasn't enabled, enabled it and ran the test again and the results are what you can see above :)
Click to expand...

Hmmm, that kind of implies that the benchmark does writing... hmmm.... I always thought HDTach was a read-only bench. How odd. Oh well, will have to check up on that then. :confused:
 
From someone on this forum:

Burst speeds on Intel boards tend to be mad due to the burst coming from the cache on the ICH chip, therefore, the max bandwidth is that of the ICH to MCH interconnect.
Click to expand...
 
You must log in or register to reply here.
Back
Top Bottom