A8N32-SLI RAID Flakiness

[Demigod's Owner]

I ve got the nforce4 x16 mobo from asus. I attached two serial ata 2.0 (or 3gbps or whatever they are calling it this week) seagate 7200.9 drives to it. Configured it first as RAID 0, benchmarked using IOMeter and HDTach, then in RAID 1, then as single drives.

Before that tho, I used an old Intel 845PEBT2 motherboard with the now old Silicon 3112(?) RAID controller, which only supported Sata 150.

I am attaching a screenshot of the HDTach scores for the two drives in RAID 1. The blue graph is the Intel Mobo, the Red Graph is the Asus. The results for the RAID 0 mode are even more fluctuous (may not be an actual word).

I am using Motherboard Bios version 1009, chipset driver 6.85 and the NVRaid controller is using the Nvidia driver. I ve heard reports about windows version of the driver being better, but nothin was auto-installed for me when I chose that option. So I went with this.

My question is, has anyone else been seeing these sort of, all over the place, results ??

Of course, any suggestions on how to fix it would be awesome also.

 

[Demigod's Owner]

I did some more reading on RAID. It seems a correct implementation of RAID 1 is supposed to boost read performance as well (not the case in my Older Intel mobo, tho I had hoped it might be in this new Nvidia one).

Additionally, since all the data exists in two or more copies, each with its own hardware, the read performance goes up roughly as a linear multiple of the number of copies. That is, a RAID 1 array of three drives can be reading in three different places at the same time. To maximize performance benefits of RAID 1, independent disk controllers are recommended, one for each disk.

courtesy of Wikipedia.

Nvidia's documentation for their RAID solution on the Nforce4 x16 platform, makes no mention at all of increase in read speeds (whereas it boasts about gains in read& write for RAID 0).

Is this sort of a "redundancy only" RAID 1 typical for integrated solutions? The HDTach and I/OMeter benchmarks for my Seagate drives in RAID 1 and single drive showed no gains in read speeds (the only discernible difference was a doubling of I/O operations while in RAID1).

 

[unhappy_mage]

"Correct" is the wrong term for that; that would be a "optimal" implementation of raid 1. All raid 1 means is your data is written to both disks and read from one or both. It's better for performance to read from both, of course, but nothing guarantees that.

 

[general]

Keep in mind also that if you use RAID-0 as your only storage, you're putting your data at a serious risk. If one of the drives goes bad or they get corrupted kiss ALL of your data goodbye. In such a case, you'd be better off getting a Raptor for your main drive, mirroring (RAID-1) those two discs and just backing up an image of your OS drive onto it once every so often.

 

[just2cool]

Yeah, RAID0 is risky. But he's not really asking for that advice. I mean, I've been running a 3x raptor raid0 for almost 2 years now without any hiccups whatsoever. You can tell me it's risky, but I already know it is.

...however, I plan to buy one of those new 750GB drives once the price comes down so I can backup my important data lol.

Anyway, those results look really strange. However, HDTach isn't the best disk benchmark anymore. Try using ATTO instead, and post those results.

You can get ATTO here: http://www.short-media.com/download.php?d=108

 

[Demigod's Owner]

I ran ATTO twice, to see if there were any noteable changes. The second run the 32mb Read speeds were better (in the mid 60s).

Since I am not familiar with ATTO, I am posting the results as someone suggested, with the hopes that someone can explain it to me:

Basically all I got out of it is that these results don't look right (compared to what the developer of the software has showing for RAID1).

I mean if I opt to go the windows XP Software RAID 1 Route, what am I looking at in terms of performance? I mean what sort of CPU performance sacrifice in return for what sort of read performance boost?(if any)

 

[just2cool]

Here are my results for RAID0:

I'm also using an nforce based board; an nforce3 ultra. You seem to have a random dropoff point just like I do. I've tried to get answers for it but nobody knows. Also, no one knows why my write speeds are almost always faster or equal to my read speeds, which is your case as well.

So, I'd say that you're running normal for an nforce based board, based on my results being similar to yours.

 

[Demigod's Owner]

Just2cool, couple of things I noticed. It looks like when you are benching with ATTO, you are choosing the "Neither" option instead of "Overlapped I/O". As I am not familiar with this software, I don't know what, if any, significance that has.

Also, since I am in RAID 1 and you are in RAID 0, in my case, the lack of improvement in write speeds is expected, in fact I am surprised that it looks as if there is almost a doubling of read speeds. I am surprised because when I use HDTach and compare single drive read speeds to RAID 1 read speeds, there is no gain (none at all). Whereas in "optimal" implementation of RAID 1 I should get almost double the read speeds.The surprise is due to HDtach saying one thing and ATTO saying another. In your case though, with RAID 0, you should not have such a disparity. Perhaps the block size you chose when building the RAID array creates too much overhead during writes? ( I am using 16k blocks). When formatting the drive in windows I also chose 16k block size.

I am now considering an alternative approach, bypassing the NVRaid controller altogether and using windows software RAID 1. I will run benchmarks on speeds and number of I/O operations later, but I am really concerned about the CPU utilization. If anyone has any suggestions, warnings etc. would be appreciated.

 

[Demigod's Owner]

I just found out that I cannot do Software RAID 1 in Windows XP Professional... Only Software RAID 0.

It seems I will need win2k Server for that stuff. If anyone knows of a ROBUST third-party software RAID1 solution, please let me know.