Poor Software Raid Performance, OS X

[Terrabit]

Hi.,

I was curious about the performance software RAID. So i compiled Bonnie++ hard drive benchmarking tool and installed OS X on the following 6 drive configurations:

* 1x Drive
* 2x Drive Mirror (RAID-1)
* 2x Drive Stripped (RAID-0)
* 3x Drive Stripped (RAID-0)
* 4x Drive Stripped (RAID-0)
* 4x Drive Stripe of Mirrors (RAID-10)

The results are quite disspointing for READ operations, particularly on Mirrored Drives

http://macprojournal.com/soft-raid.html

This should be interesting reading for anybody considering a RAID setup on the Mac Pro.

Terrabit.

 

[Apple2Mac]

It's likely your Seagate drives, I've read some very disappointing results with Seagate even in Hardware RAIDs
Take a look at this bare feats test http://barefeats.com/quad08.html
That Seagate drive you used either comes in second or dead last, there is something spotty about a that drive....

 

[Altimeter88]

I have 4x500GB in RAID 10. They are all the WD drives (WD5000KS) which are supposed to be the quietest and best performing of the 500GB drives available. I would love to run the benchmark on my new Mac Pro. It is also 2.66Ghz but has 4GB of RAM. I don't know how to use that source code program so if you know of another utility or can tell me where I can get a compiled version that I can just install that would be great.

 

[Origin]

It's very interesting. I currently have a RAID0 Mac Os X partition with 2x200 GB SpinSpoint Samsung drives and I wondered if the RAID1 would be as fast as my setup for common reading access because I don't really need 400 GB on my system drive.

 

[seanf]

Seagate drives ship in SATA 150 mode and need a jumper removing to operate in full SATA 300 mode - see here for details. Could this be the cause of your problem?

Sean

 

[patseguin]

What's the differrence between RAID0 and RAID1? My MacPro has a 250GB Seagate drive and I ws thinking of buying anohter 250GB drive for a RAID setup.

 

[andrewfee]

What's the differrence between RAID0 and RAID1? My MacPro has a 250GB Seagate drive and I ws thinking of buying anohter 250GB drive for a RAID setup.

RAID0 spreads the data between two drives evenly, and the computer will see it as one 500GB drive. This should improve both read and write performance, if I remember correctly. If one of the drives fail, I believe all the data on both drives is lost.

RAID1 would show up as one 250GB drive, and sends identical data to both. This can improve read performance, and greatly improves reliability - if one drive fails, the data is stored on the second. You can then replace the faulty drive and rebuild the array. Say you had a 500:1 chance of one drive failing and you losing data, RAID1 increases your odds of losing the data to 250,000:1

 

[Origin]

RAID0 = Stripping. No security, the aggregated volume is two times the size of the smallest drive involved in the RAID setup. For example, a 2x200 GB RAID0 will provide you 400 GB of unescured space. The main advantage is that you get your data stored on all the drives in stripes, so you get a great performance boost, specially in read opérations, because your operating system access the two drives synchronously, combining the speed of two (not in all cases but very often).

RAID1 = Mirror. Full security, the two (or more pairs) drives are combined to be written synchronously with the same data. When you write a file on the mirror, it is copied on the two drives. The benefit is that you can loose one of the drives without loosing your data. The counterpart is that your space is divided by 2 compared to the raw capacity of all of the drives involved in the mirror. For example : 2x200 GB RAID1 provides you 200 GB of full secured space.

 

[patseguin]

Which config is commonly used to boost performance? I didn't (yet) read the original poster's review but I am interested in how well OS X does with software RAID.

 

[Origin]

Just to give you an example : my MacPro setup use a 400 GIG RAID0 composed by two 200 GB Samsung SpinPoint .... and the MacPro's booting (from the grey apple logo to the desktop and all my menuextras loaded) in less then 6/7 seconds !

I spend more time typing my passowrd than the desktop to be completly operationnal

The performance of the Software RAID0 on theses drives is awesome !

 

[mkrishnan]

Thank you for sharing all this information!

I'm not technically proficient on disk benchmarking... so I'm sorry if I say something stupid. But I remember that there were initially issues with Intel-Based Macs and benchmarking apps that had to do with some subtlety of the system... since you've compiled a program yourself and are one of the few people using this benchmarking tool on OS X... are you sure you believe your results? Is there any way to do a control condition in which you know what the answer should be? (e.g. if you benchmark the non-RAID MacHFS+ boot partition, do you know what to expect? In this way, this benchmark would serve as a calibration of all the other benchmarks you did).

 

[ac4lt]

Hi.,

I was curious about the performance software RAID. So i compiled Bonnie++ hard drive benchmarking tool and installed OS X on the following 6 drive configurations:

[snip]

The results are quite disspointing for READ operations, particularly on Mirrored Drives

Actually, your results looked pretty reasonable, at least in their relative differences. I think your expecting more from RAID-1 read performance than you should. The reason RAID-0 read performance is so good is that only half the data is on each drive. With RAID-1, all the data is on each drive. Only one drive is used for reading. I would expect RAID-1 read performance to be essentially identical to a single drive's read performance. The fact that you got slightly better performance surprises me a bit (though it may be test variability).

That's the whole reason RAID-10 came about. You got the read and write performance of RAID-0, the protection of RAID-1 and without the write performance hit of RAID-5. The trade-off is space efficiency since you lose half your space to mirroring.

Anyway, leaving the absolute numbers you got aside since I can't really judge them, the relative differences in your results look pretty much like I would expect.

 

[Abulia]

That's the whole reason RAID-10 came about. You got the read and write performance of RAID-0, the protection of RAID-1 and without the write performance hit of RAID-5. The trade-off is space efficiency since you lose half your space to mirroring.

How does that compare to RAID 0+1? (Does the Mac even support 0+1?)

I too wasn't really surprised by the reults; pretty much what I would expect.