Sign in with
Sign up | Sign in
Your question

Strange HD Tach graph or normal?

Last response: in Storage
Share
May 4, 2007 7:39:53 PM

This is the drive where my OS & Page File are located using a 3ware RAID controller with the system described below. I'm very new to using HD Tach and something tells me the graph shouldn't have so many peaks and valleys or is this normal? Does this look right to you guys?





Vista 32bit | Core 2 Duo E6600 | BFG GeForce 8800 GTX | Patriot EP 2X1GB PC2-8500 DDR2-1066 CL5-5-5-9 | 3ware 9650SE-4LPML RAID Controller with BBU | 150GB Raptor x2 RAID 0 - OS/Swap | 150GB Raptor x2 RAID 0 - Data | 150GB Raptor - Backup | SB X-Fi XtremeGamer Fatal1ty Pro Series | Dell E207WFP & Samsung SyncMaster 213T | Asus Striker Extreme | PC Power & Cooling Turbo-Cool 1KW-SR | Silverstone Temjin TJ07 | Zalman CNPS9700 NT | 3DMark06: 9705
May 13, 2007 12:44:24 AM

Did you run a Long benchmark? It's the only one that really counts...

The average of 131MB/s sounds about right for a pair of Raptors.

Here's my HDTach set to run a Long benchmark on a 4 drive RAID10, which is identical to what it was in a 2 drive RAID0 before I got the other 2.

RAID10 HDTach. An average of 92MB/s for Seagate 7200.10s. Your raptors are about 40MB/s higher, as expected.

Just for fun: RAID5 HDTach. RAID5 appears to be very good at HDTach boasting 192MB/s average. Both are from the same set of drives, using Matrix RAID.
May 13, 2007 12:50:25 AM

I ran the quick bench - 8MB. I was just worried that the graph looked strange. I'll make sure to run the long benchmark next time. What's the significant difference between the two?
Related resources
May 13, 2007 3:01:28 AM

The 8MB benchmark isn't as accurate. It'll end up with, well, with big spikes and uneven looking curves.

The 32MB benchmark should be smoother, like the RAID5 image. I'm a little curious myself why the RAID10 looks as it does. If you excuse my geekness, it looks like gaussian wave packets occuring at a regular period...
May 13, 2007 3:50:40 AM

I've found HD Tach's results to be very bizarre with any sort of RAID, regardless of whether you use 8mb, 32mb or variable zone tests.

The only thing I don't really know is, what would be the best tool to use for benchmarking RAID setups.

-Brad
May 13, 2007 1:02:10 PM

Here is the result using 32MB zones - long bench. 9650SE hardware RAID controller with 2 150GB Raptors. What do all the peaks and valleys signify?

May 13, 2007 4:00:08 PM

It signifies HD Tach's uselessness for benching RAID controllers.

Leave a post on the publisher's web site and see if they answer you.

Don't hold your breath.


-Brad
May 13, 2007 4:51:23 PM

The makers of HD Tach have a forum with plenty of RAID/raptor and questionable performance issues. I'll go raise heck there :D 
May 13, 2007 5:20:53 PM

I disagree slightly with HDTach being useless for RAID. I can't disagree completly, though.

Look at my RAID10 -- it's useless. The RAID5, that's not useless, in fact it looks normal. Your RAID0 Raptor setup gives a fairly useless result, unless you take the peaks and valleys and treat it as a sinusodial wave, and just take the average value...that is, just extend the end of the curve back to the beginning as a straight line through the center of the peaks and valleys. Of course, this introduces an element of error, but I suppose it wouldn't be much of a skew, if we assume the end of the graph to be accurate.

Now, I also had RAID0 in place of RAID10 and it looked the same. And my RAID0 on a 2nd PC looks normal.

The only statement I can make based on that is that some RAID0 setups, or RAID setups with RAID0 elements, do not perform as expected under HDTach. It might depend on the controller. The RAID10 is an ICH8R (as is the RAID5), while the normal RAID0 is an nForce 2 (running Seagate 7200.7s 2x120GB).
May 13, 2007 6:19:03 PM

I have found that when I am setting up my raid drive and set the block size small it makes hd tach numbers go up considerbly. But I have experimented with 2 3 and 4 drives in raid zero the synthetic benchmarks never seen to carry over in any real world use.
try loading a game with raid 0 then with one drive, you really dont see a difference, only difference you will see is when you loose all your data do to your raid hiccuping or a drive failing.

But try putting your block size down to 8k or 16k you will see those hd tach scores go though the roof
May 13, 2007 6:26:34 PM

When I installed and setup the RAID controller, the controller's default stripe size was 64KB. I went with that because I'm still in the learning phase of all this. Through threads like this, I get to learn a little more. I might experiment with different stripe sizes and see what happens. Will lower values increase real-world performance? I want to try it, but it'll have to wait until next week. Thanks to all in the interim :) 
!