RAID benchmarking

betaflame

Member
Jul 28, 2009
81
0
0
I'm building some computer for a small company (computer game/hardware dev) and I have on hand an X58 ICH10R Gigabyte board with the Jmicron controller as well.

I thought I'd run a few benches of my own on the following

ICH10R:
RAID0
RAID1
RAID5

Windows Server 2008 Software RAID:
0
1
5

Jmicron Controller on the Gigabyte board:
0 and 1

HDTach/HDtune work well for the hardware raids, but I have yet to find a program that can use the logical drives generated by Windows Software RAID.

Iometer I think can do it, but it has an absurd learning curve (spent an hour to no avail).

Anyway, does anyone have any way to make HDTune/HDTach work, or do you know of a program that can bench logical drives properly.

(The RAID will be empty so I can do destructive write tests)
 

Nothinman

Elite Member
Sep 14, 2001
30,672
0
0
Frankly I'd consider any benchmark that doesn't go through the filesystem to be pretty much worthless since all of the I/O done on the system will be done through the filesystem. They'll give you an idea of the the disk max throughput but nothing realistic.

And while I highly doubt the ICH10R RAID will be faster, I'd consider the flexibility of software RAID to kill any minor performance increase you might see from that controller.
 

betaflame

Member
Jul 28, 2009
81
0
0
Well, I know all 3 use the CPU with their own set of drivers and features:
Intel and Jmicron are pseudo-hardware raid, and windows is pure software.

My basic assumption is that it's going to play out
ICH10R>Jmicron>Windows in all the tests, but I've seen some crazy anomalies in the past (partition/stripe alignment and such)
 

Nothinman

Elite Member
Sep 14, 2001
30,672
0
0
Intel and Jmicron are pseudo-hardware raid, and windows is pure software.

The "pseudo" is the important part because it kills the whole "hardware RAID" thing. The only thing the chipset does is some magic to make them bootable, all of the real work is done in the driver so you end up with similar CPU usage and less flexibility.

My basic assumption is that it's going to play out
ICH10R>Jmicron>Windows in all the tests, but I've seen some crazy anomalies in the past (partition/stripe alignment and such)

I can't speak to Windows software RAID too much but Linux software RAID beats out most hardware controllers handily. The only reason to choose hardware RAID over Linux software RAID is if you've got a hardware XOR engine, battery backed cache, etc.
 

betaflame

Member
Jul 28, 2009
81
0
0
Correct. The performance difference comes from the drivers. The advantage of the psuedo hardware is that is interoperable with all current OSes, and is not dependent on having windows manage the raid (a thought which makes me slightly uncomfortable).
 

Emulex

Diamond Member
Jan 28, 2001
9,759
1
71
i bet the raid-5 will suck so much butt its not funny.

without a bbwc you will hit stupid penalty for the double reader and writing to all drives
 

Nothinman

Elite Member
Sep 14, 2001
30,672
0
0
The advantage of the psuedo hardware is that is interoperable with all current OSes, and is not dependent on having windows manage the raid (a thought which makes me slightly uncomfortable).

That's a disadvantage because if that board gets replaced you'll have a some fun getting your array back. With software RAID you just plugin the drives and import the array.
 

betaflame

Member
Jul 28, 2009
81
0
0
I know for a fact you can plug an ICHXR generated raid into any ICH8R/ICH9R/ICH10R and it will appear normally (the raid configuration data is stored on the drives themselves)

The Jmicron I have no idea but I assume it would be similar.

Yes software is the most movable, but it's dependent on the OS for much much more than just standard filesystem read/write.
 

Nothinman

Elite Member
Sep 14, 2001
30,672
0
0
The Jmicron I have no idea but I assume it would be similar.

I wouldn't assume anything with those cheap, onboard controllers.

Yes software is the most movable, but it's dependent on the OS for much much more than just standard filesystem read/write.

What "much much more"? The only difference between the onboard software RAID and Windows software RAID is essentially where the RAID information is stored. In the former it's (IIRC) at the end of the disk and outside of the OS's reach and in the latter (with Windows anyway) it's part of the Dynamic Disk configuration.

The only major downside to using Windows software RAID is the fact that lots of partitioning tools don't understand Dynamic Disks. But that's easily avoided.
 

betaflame

Member
Jul 28, 2009
81
0
0
You're right, I forgot how annoying dynamic disks were. Had some issues with them before.

By rely on the OS I meant you can't shift the raid over to linux or something if you decide you want to. Also if something goes horribly wrong in the OS it's feasible it could hose the the RAID configuration.

If that happened with an ICH10R version at worst you'll get some bad data written to the drive, the raid config is isolated (I guess that's the point I'm trying to make)

Linux raid is the best software raid I know, but that's been benchmarked to death already.

Edit: Crystalmark works but it doesn't account for position on the drive. Shouldn't matter as all the arrays are empty. I'll use HDtune/HDtach as well on the ICH10R and Jmicron.
 

Nothinman

Elite Member
Sep 14, 2001
30,672
0
0
You're right, I forgot how annoying dynamic disks were. Had some issues with them before.

They're not that bad as long as you remember they're there and that crap like old Partition Magic won't know what to do with them.

By rely on the OS I meant you can't shift the raid over to linux or something if you decide you want to. Also if something goes horribly wrong in the OS it's feasible it could hose the the RAID configuration.

Actually Linux can read Dynamic Disks so it would probably work, but you can't really shift over to Linux without converting from NTFS to ext3, XFS, etc either so I don't see that as a problem. And if something goes horribly wrong with the OS it's just as likely to host your filesystem.

If that happened with an ICH10R version at worst you'll get some bad data written to the drive, the raid config is isolated (I guess that's the point I'm trying to make)

Not necessarily. The driver has to have access to the RAID config since you can use that Intel Matrix crap in Windows to mess with the config so it's possible that it can screw that up too. And if you've got bad data coming from somewhere and headed toward your drive chances are it's going to affect either the data in the filesystem or the filesystem itself and the RAID format won't matter.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
Originally posted by: betaflame
Edit: Crystalmark works but it doesn't account for position on the drive. Shouldn't matter as all the arrays are empty. I'll use HDtune/HDtach as well on the ICH10R and Jmicron.

Yeah its not perfect but it is an easy way to get those "nice to see" 4KB random write results.

So too the AS SSD benchmark, check it out if you have the time and inclination.