ICH10R vs JMB36X vs Windows software raid

[betaflame]

I was fairly bored so I took 3 WD10EAVS (1TB 5400RPM Green) drives and did assorted RAID configurations with them for benchmarking. The results are interesting, with unknown limitations coming out of the woodwork. HDtune and any block level benchmark will not work on windows software raid, ergo crystaldiskmark.

Write back and drive caches are always enabled.

1x WD10EAVS:
http://img91.imageshack.us/img91/284/1x1tbsimple.jpg
http://img30.imageshack.us/img...ebenchmarkwdcwd10e.png

2x 1TB RAID0:
notes:
-Jmicron's bus speed is limited to 1.5Gb/s total (the entire controller, thus the terrible performance)
-Windows performs admirably, but it's incredibly difficult to install windows on a windows software RAID0

ICH10R:
http://img504.imageshack.us/im...ebenchmarkintelrai.png
http://img512.imageshack.us/im...88/2x1tbraid0intel.jpg
a respectable showing by intel. NCQ and write-back cache help.

Windows Software:
http://img23.imageshack.us/img...100/2x1tbraid0soft.jpg
An excellent showing by Windows, beating out intel by a few megabytes. Also the 512K writes get combined by windows, leading to a massive performance gain.

JMB36X:
http://img14.imageshack.us/img.../2x1tbraid0jmicron.jpg
http://img30.imageshack.us/img...hdtuneraid0jmicron.jpg
Limited by its bus, this controller performs terribly.

2x 1TB RAID1:
notes:
-Windows ONLY READS OFF ONE DRIVE IN RAID1 (that's very important). Intel and JMB read off both (for error checking)

ICH10R:
http://img38.imageshack.us/img...70/2x1tbraid1intel.jpg
http://img512.imageshack.us/im...benchmarkintelraiw.png
-A good showing, just under Single drive speed

Windows Software:
http://img22.imageshack.us/img...720/2x1tbraid1soft.jpg
-The same speed as one drive on read as it only reads off one drive. Writes to both, quite nice. A very good solution for insulating again drive failure, no error checking.

JMB36X:
http://img8.imageshack.us/img8...ebenchmarkezbackup.png
A horrible showing. You should use windows RAID1 if you can.

3x 1TB RAID5:

ICH10R:
http://img14.imageshack.us/img...benchmarkintelraif.png
http://img504.imageshack.us/im...69/3x1tbraid5intel.jpg
A strong showing, stable all around. Good performance, just under a RAID0 of 2 drives.

Windows Software:
http://img219.imageshack.us/im...7808/3xtbraid5soft.jpg
A good showing on read speeds, staying near intel. Unfortunately it gets destroyed on the writes. Probably a function of Intel's emulated write-back cache and pseudo-hardware controller

Conclusions
1)Don't use the Jmicron controller for anything other than dumb AHCI, preferably only for DVD drives.
2)Intel is better for RAID0 on an OS array, RAID5 in general, or on any multi OS system. Also movable to any other ICHXR systems.
3)Windows has very nice RAID1 and all arrays are movable between any Windows Vista and newer systems. (Dynamic partitions can only be read by Windows)

That's all, any questions I'll answer.

 

[taltamir]

1)Don't use the Jmicron controller for anything other than dumb AHCI, preferably only for DVD drives.

I don't think I have seen someone say something nice about jmicron in any of their market segments yet...

 

[ch33zw1z]

Originally posted by: taltamir

1)Don't use the Jmicron controller for anything other than dumb AHCI, preferably only for DVD drives.

I don't think I have seen someone say something nice about jmicron in any of their market segments yet...

JMicron JMB363, worst IDE/SATA/eSATA (which windows see's as a scsi device) ever. Disable it and find any other solution. I battled a couple weird IDE CD/DVDROM issues for about a year until I disabled it and just bout a SATA drive.

Coincidentally, I was again(2nd time) trying to update the firmware(currently at 1.06.69) on the JMB363 onboard an Asus P5K, it repeatedly fails with "unsupported flash" or "no flash found"...no matter what I try to do, the results stay the same. Looking through the Readme.txt that comes with the firmware update utility, it only supports certain flash. So I can only assume that the flash used for this setup is not supported by JMicron.

 

[taltamir]

jmicron IDE/SATA/eSATA is used for two SATA ports on my mobo and ide... 6 ports come from the intel ICH10R (or was it 9R?)... well. If I do not disable the jmicron controller, windows is liable to get stuck when I attempt to go into sleep mode. luckily i don't use IDE and the 6 ports from the southbridge are enough for me, so I just disabled it.

 

[Golgatha]

Not to thread jack, but this is something I need to decide on. I have a friend who wants me to setup a basic home office for them. Your post made me realize I have some choices.

I'm resolute on using two 500GB RE3 Western Digital drives in RAID1 to provide redundant backups, as well as using Acronis to image the RAID1 to a physically separate 500GB RE3 drive each night. I think to rebuild the machine (if ever necessary) it would be nice to not have the OS install on the RAID1 array. I'm thinking of doing it this way.

100GB partition for OS and apps on single RE3 drive.

400GB space for Acronis image of RAID1 array.

500GB RAID1 array to share out to users and map their My Documents, etc. to it.

Looking at your benchmarks, and knowing this company will stay with Windows software, I think it might be wise to just use dynamic disks to make the RAID1 (I have the option of using the ICH10R chipset). As you stated, the data should be transferable to any system that's running WinXP Pro or later down the road correct?

 

[taltamir]

how can you image a 500GB array to a 400GB partition?

Also i thought you said the image will be on a physically seperate device, so how can it have a 100GB OS partition?

I wouldn't put windows on a RAID array either.

 

[Golgatha]

Originally posted by: taltamir
how can you image a 500GB array to a 400GB partition?

Setup a nightly imaging with the Acronis Scheduler and use sufficient compression to make sure the 500GB RAID1 will fit on the 400GB backup partition.

Also i thought you said the image will be on a physically seperate device, so how can it have a 100GB OS partition?

The single (i.e. no RAID) disk would be partitioned to 100GB and 400GB of space. 100GB partition will have the OS/apps/etc installed on there. The other two disks will be in RAID1 and that's where the 5 users will save their data to. Acronis will run on a schedule (probably around 3am or so) and make an image file of the 500GB RAID1, which will be saved to the 400GB partition on the single (i.e. no RAID) drive that also houses the OS on the 100GB partition. This way, there is a real time backup of their data and a nightly backup of the already redundantly backed up data on a physically separate drive (3 copies of the same data total). Make sense?

I wouldn't put windows on a RAID array either.

What's the disadvantage over fakeraid on the Intel ICH10R? A power failure could result in lost data with either one. I'd personally like to get a higher-end hardware RAID card to avoid this, but it's not in the budget.

 

[Golgatha]

-Windows ONLY READS OFF ONE DRIVE IN RAID1 (that's very important). Intel and JMB read off both (for error checking)

Just noticed this note. I'll be using the ICH10R apparently. This is very, very important data to them (although apparently not important enough to budget for hardware RAID).

 

[betaflame]

I should mention that the part about error checking is my assumption. As things stand, I can think of no other reason to read the same data from both drives other than checking for errors.

This does not mean it can regenerate the correct data, as it does not know which copy is right, merely that there is a problem.

It will probably bring it to your attention with the hardware error dialog. Then you would probably manually look to see which file is correct. Better than having no option.

 

[ArchAngel777]

I sincerely appreciate you posting this mini review/benchmarks. This is very useful information.

 

[Syntax Error]

RAID 1 should theoretically double read speeds as a smart RAID controller would utilize the presence of two disks to read bits in parallel, but unfortunately, this is hardly done in reality except in the more higher-end RAID controllers.

 

[RebateMonger]

Originally posted by: betaflame
As things stand, I can think of no other reason to read the same data from both drives other than checking for errors.

To speed up RAID 1 reads, some systems are able to grab data from both drives at once.

 

[betaflame]

Yes, I assumed that and if I recall that's how some of the SCSI hardware RAID controllers I've played with worked.

However, it appears as though the ICH and JMB were reading every block. Otherwise I wouldn't have seen the 90-95% performance of a single drive. It's all a bit odd.

 

[Tencntraze]

I actually seem to be having the opposite results on my end, though the numbers that I'm shown are pretty sketchy (especially the burst rates). The JMicron benchmark was done on Win7 RC with an install that had been there for at least a month, while the Intel ones were done this morning after a fresh install of Win7 RTM with the latest Intel Matrix Storage Manager:

JMicron

ICH9R No WriteBack
ICH9R WriteBack

As I mentinoed, the burst rates seem pretty absurd at points, but on the other hand I wonder why it's so much slower than the JMicron for minimum rates.

 

[betaflame]

The results look the same as mine. Just that something was accessing the ICH9R array when you were benching it.

 

[Emulex]

raid-1 should read off the drive that has the least activity. in theory an inactive system should see near double read speeds. hp's sas controllers do this.

there is no reason to read from both disks in raid-1 otherwise; since if there is no read error the data is intact.

i'd never use ich for raid-5.
hell even raid-1 unplug the power a few times and see how that raid-1 sits.

 

[betaflame]

Just got win7 ult RTM, all software RAIDs are enabled (0,1,0+1,5)

 

[Tencntraze]

Originally posted by: betaflame
The results look the same as mine. Just that something was accessing the ICH9R array when you were benching it.

I wonder what that could have been. I've tried redoing this benchmark many times with the same result for the ICH9R. Not really a big deal to me if it doesn't affect real-world performance, but am certainly curious.

 

[betaflame]

Something that occurred to me is that if you are running the JMB controller in AHCI mode, and then use windows raid1, by reading off only 1 drive, it won't saturate the bus, and you'll still have full bandwidth for a normal magnetic drive raid1, so no real penalty.

 

[steve0701]

While Win 7 is said to do a superior job controlling a pair of Raid 0 drives, is it true that Win 7 (Home Premium) itself must be installed elsewhere, such as on a 3rd drive? I've got a pair of empty HDDs and I'm looking for a step-by-step tutorial before I give up and just go with the Intel ICH10R built into the motherboard.

 

[RebateMonger]

While Win 7 is said to do a superior job controlling a pair of Raid 0 drives, is it true that Win 7 (Home Premium) itself must be installed elsewhere, such as on a 3rd drive? I've got a pair of empty HDDs and I'm looking for a step-by-step tutorial before I give up and just go with the Intel ICH10R built into the motherboard.

You can't boot to any form of striped RAID array when using Windows Software RAID. That's because Windows has to be running first to control the striping process. Windows Software RAID can boot on a RAID 1 array, but not on RAID 0 or RAID 5.

 

[VirtualLarry]

Well, count me as the odd man out, I haven't had any trouble from the Jmicron controllers. I leave them set as IDE mode, put my IDE DVD-burner, and SATA primary bootable HD on them, and then set the Intel ports to RAID mode. This allows me to install XP *without* using the F6/floppy procedure, as the OS drive is in IDE mode.

Never had any noticable performance or compatibility issues.

Friend is running off of the IDE port on an IP35-E, with an IDE Seagate 250GB HD, and an IDE DVD-burner. System runs fine, except for overclocking the PCI-E bus and attempting to burn DVDs. Then the Jmicron doesn't like that.

Edit: It should be said that I always leave the Jmicron ports in IDE mode for the SATA ports, and I always use the MS driver, never the Jmicron driver.

 

[Emulex]

You can't boot to any form of striped RAID array when using Windows Software RAID. That's because Windows has to be running first to control the striping process. Windows Software RAID can boot on a RAID 1 array, but not on RAID 0 or RAID 5.

My first knowledge of using consumer drives in a raid scenario was this - raid 1 software. windows like intel matrix will crap your raid and there is NO WAY to flip the switch back to "ok" to boot once it fails the raid (fail is worst then degraded).

 

[r5a]

Thanks for doing this, found from Google!

I have a evga x58 with these two exact controllers.

I can also vouch raiding two really good drives in R0 is absolutely terrible with JMicron. Just backing up now and moving over to the ICH10R ports.