Can Mac Pro do software RAID 0 for the systems disk ?

[aponsin]

I do manipulate large images files (~1 GB) and in order to make the whole process faster I know that I need really fast hard drive... hence the RAID.

But I can't afford to by the hardware RAID card... so can I have my macosx install and my Photoshop scratch disk be installed on a twin HD in software RAID 0 ?

How would I go in doing that if it's possible ?

Alex

 

[ksz]

Yes you can absolutely use software RAID-0 on your boot drives. I'm doing this with my Mac Pro (early 2008 model).

The steps I followed to do this:

1. Inserted 2 identical drives in internal bays 2 and 3.
2. Left the original disk in bay 1 and booted off the original disk.
3. Started Disk Utility (in the Applications --> Utilities folder).
4. Setup RAID-0 on the disks in bays 2 and 3 and formatted/mounted the new filesystem. Give this volume a unique name.
5. Downloaded Carbon Copy Cloner.
6. Copied the entire original disk to the new volume using CCC.
7. From System Preferences --> Startup Disk, set the new disk to be the Startup Disk.

That's it. However, I cleaned things up a bit as follows:

8. Opened the Mac Pro case and moved the two RAID-0 drives into Bays 1 and 2. Moved the original disk into Bay 3 (you could have done this at the very start).

9. Used the original disk (now in Bay 3) for Time Machine.

 

[NoNameBrand]

Why do you want a fast boot drive? don't you want a fast scratch disk and fast data disk?

If it were me, I'd make the boot volume RAID1 or a single disk. Make the scratch disk as fast as you can (3-4 multi-disk RAID0, from partitions at the start of the disk), and larger partitions on the same disks in a RAID0 for data - Photoshop blocks on opening/saving, so you're not going to be doing a lot of I/O on the same disks at the same time.

Make sure you backup your data, though, due to the dangers of striping.

 

[aponsin]

Why do you want a fast boot drive? don't you want a fast scratch disk and fast data disk?

If it were me, I'd make the boot volume RAID1 or a single disk. Make the scratch disk as fast as you can (3-4 multi-disk RAID0, from partitions at the start of the disk), and larger partitions on the same disks in a RAID0 for data - Photoshop blocks on opening/saving, so you're not going to be doing a lot of I/O on the same disks at the same time.

Make sure you backup your data, though, due to the dangers of striping.

I actually would like both LOL, I am also thinking of moving to hardware RAID later.. I just can't afford it for now... I am also considering an SSD for the boot drive...

Anyway thanks a lot for the process description, makes perfect thanks !!

Alex

 

[NRose8989]

Why do you want a fast boot drive? don't you want a fast scratch disk and fast data disk?

If it were me, I'd make the boot volume RAID1 or a single disk. Make the scratch disk as fast as you can (3-4 multi-disk RAID0, from partitions at the start of the disk), and larger partitions on the same disks in a RAID0 for data - Photoshop blocks on opening/saving, so you're not going to be doing a lot of I/O on the same disks at the same time.

Make sure you backup your data, though, due to the dangers of striping.

Let me get this straight...

Are you saying first build a raid 0 (2-4 disks) then partition that array into 2 partitions so that the OS sees 2 independent disks?

Would there even be a performance gain from doing so?

The disks are still having to switch between the partitions anyways so theoretically wouldn't it be the same as building a single huge array and using it as your boot drive and scratch disk?

 

[EdbBob]

I have a Mac Pro 2008, and I'm running OSX from a 4x500GB software Raid0. I used Disk Utility during OSX installation to generate the raid, and then picked the raid for OSX installtion. Very easy!

Afterwords I measured about 270 MB/s, and that really helps everything I do. It's blazingly fast! And I paid something like 250$ for the 2TB raid. Beat that...

Be aware that running with Raid0 as system disk can be hazardous, as you get 4 times higher risk of a problem with the disk, but if you use Time Machine for backup, it shouldn't be a huge problem.

If you want to practice OSX installation on a raid, you can start out by installing on a raid made from two (min 4GB) USB memory keys.

You can even use Time Machine to back up your current system, and then restore that to the raid.

I love my

 

[Tesselator]

I get about 400MB/sec off my 3-drive RAID-0 and I'm using it for booting. Everything and I really mean everything is much much faster. My 8-core 2.66 MacPro from 2006 profiles very much like a 2008 2.8GHz system! Everything except number crunching! For example my X-Bench scores shot WAY up.

If you want to maximize the responsiveness of your system I highly recommend it!

If you feel like you need your boot to be a separate entity you can additionally partition the RAID set. Mine uses the first 10% or so as the "Mackintosh HD" boot, the next 70% is used for video editing (large data needing speed), and the last bit is used for storage where I don't care so much about speed. I have my MP3 files and my PDF e-books stored on the last part.

I don't remember the profile curve exactly but the 1st 15% or so of my RAID 0 is 390MB/s ~ 420MB/s. The middle and largest section tests at 320MB/s ~ 390MB/s and the slow inner sectors are like 250MB/s ~ 320MB/s. (Sustained R/W no cache.) So where you place your partitions and what purpose you allocate to them is somewhat meaningful. The drives singly get about 65MB/s average.

Also different drives preform differently in RAID than they do singly. A drive that may not be the fastest used singly may be very much faster than "the fastest" comparable drive when used in an multi-drive RAID array.

.

 

[PowerPaw]

I set mine up as RAID0 with partitions. The Mac Disk Utiltiy is a bit quirky if you have used hardware or for that matter Windows as the setup is back to front in comparisan.

Create a partition and then create the stripe, not create a stripe and then partition

The reason for partitioning is I wanted to keep my data and system independant from one another for migrating off later - makes things a little simpler. As for the location of data on the platters; well if you analyse your disk you will notice Mac OS keeps the system files together at the beginning so I didn't do it for that. It should work the same as it would for an unpartitioned system and data stripe. Time Machine should cover your butt if the stripe gives up on you. I back the system to a separate bootable firewire drive.

Over the coming months and year I'll move system to an SSD and possibly the data leaving a nice fast scratch partition. Its a really cheap way of getting good disk throughput performance which you can migrate off from as needs and finances allow. The 10K drives will give you considerably better seek times; SSDs however will blow away spindle technology in this space.

 

[NoNameBrand]

Are you saying first build a raid 0 (2-4 disks) then partition that array into 2 partitions so that the OS sees 2 independent disks?

Other way around - partition the stripes, then make separate arrays. Sure, creating a striped array on disks used for nothing else provides a greater boost than doing multiple things with them, but depending on their use, you can still see a major speed up. A huge chunk of my disks are used for backups - backups run when I'm not using the computer, so there's no competition for disk I/O.

My G5 was had a scratch array striped across multiple disks, using only a small partition from each. It provided a boost in Photoshop performance.

When my new WD Black drives arrive for my new Mac Pro, I'll be taking 15GB off the front of each disk for a striped array. The rest of the space on each disk (a - d) will be used for a) Boot/Applications + Backup of drives b and c; b) User Data, Backup of drive a; c) Games, Media, Windows crap; d) Time Machine

Would there even be a performance gain from doing so?

Depends on your usage. If a given disk is being intensively accessed for multiple things at once, then the gain won't be large. If your Photoshop scratch disk is being hammered, what else is your computer doing with those disks? (The answer may or may not be 'nothing' for you. It's typically 'not much' for me)

If your usage is 'launch app' -> 'load files from disk' -> 'do intensive things' -> 'save', then what should you optimise?
You're reading the application onceRAID1 makes more sense for the boot volume than RAID0, if you've split your data off it - but really, how long does it take a Mac Pro to launch your app? after the first time it's probably in RAM anyway. For me, my Mac is on 24/7 and I have enough RAM that there's a good chance my app is either running or cached.

You probably want to optimise the other steps. If you're using Photoshop, you aren't loading/saving while doing anything else, so your optimisations can be at odds.

Read this:
http://macperformanceguide.com/OptimizingPhotoshop-Configuration.html


Like PowerPaw, I use partitions for ease of migration and backup. Performance was secondary. My G5's partitions aren't in the right spots for the best performance, the Mac Pro's will be much better planned (partitions at the start of the disk are faster than those at the back).

 

[ksz]

I get about 400MB/sec off my 3-drive RAID-0 and I'm using it for booting.

That's pretty amazing, but I'm a bit suspicious.

1. Are your 3 drives installed inside the Mac or in an external enclosure?

2. What make/model drives are you using?

3. Are you using SAS or SATA drives?

4. If the drives are installed externally, what enclosure and controller are you using?

5. Did you use "AJA Kona System Test" to measure Read/Write speeds?

 

[noushy]

400mb/s off 3 drives?

I have to agree with the above. I ran a 4 drive Raid5 on a dedicated $700 raid controller and got 300mb/s. On Raid 0 off 4 drives got around 360mb/s and these are fast seagate 1tb drives. Running software raid 0 on 3 WD 2tb drives, I get around 250/270mb/s write/read, and these drives are considered fast for standard sata drives (not as fast as black series, but high aureal density makes them quick). 400 Mb/s off 3 drives is almost impossible, especially using software raid. I could see that being reached with 3 velociraptors, or some 10k or 15k sas drives, but sata 7200 rpm drives, no way. Please give more information.

Peace,
Noushy

 

[Tesselator]

That's pretty amazing, but I'm a bit suspicious.

Why suspicious?
EDIT: Nevermind, I just read noushy's post.

1. Are your 3 drives installed inside the Mac or in an external enclosure?

2. What make/model drives are you using?

3. Are you using SAS or SATA drives?

4. If the drives are installed externally, what enclosure and controller are you using?

5. Did you use "AJA Kona System Test" to measure Read/Write speeds?

1. Yes, internal.
   
   

2.   Model:			Maxtor 7V300F0
     Revision:			VA111900
     Serial Number:		V60QWG9G
     Native Command Queuing:	Yes
     Queue Depth:			32
     Removable Media: 		No
     Detachable Drive:		No
     BSD Name:			disk1
     Bay Name:			"Bay 2"
     Mac OS 9 Drivers:		No
     Partition Map Type:		GPT (GUID Partition Table)
     S.M.A.R.T. status:		Verified
   _______________________________________________
   
     Model:			Maxtor 7V300F0
     Revision:			VA111900
     Serial Number:		V60MAJ7G
     Native Command Queuing:	Yes
     Queue Depth:			32
     Removable Media:		No
     Detachable Drive:		No
     BSD Name:			disk2
     Bay Name:			"Bay 3"
     Mac OS 9 Drivers:		No
     Partition Map Type:		GPT (GUID Partition Table)
     S.M.A.R.T. status:		Verified
   _______________________________________________
   
     Model:			Maxtor 7V300F0
     Revision:			VA111900
     Serial Number:		V60QP24G
     Native Command Queuing:	Yes
     Queue Depth:			32
     Removable Media:		No
     Detachable Drive:		No
     BSD Name:			disk3
     Bay Name:			"Bay 4"
     Mac OS 9 Drivers:		No
     Partition Map Type:		GPT (GUID Partition Table)
     S.M.A.R.T. status:		Verified

3. SATA.
   
   
4. N/A
   
   
5. I used 5 or 6 different speed tests and everytime I find a new one I run it too.
   I can search my HD and tell you the names if you're interested. Spotlight with
   system files included doesn't turn up any apps with "AJA" or "Kona" in the name
   so I guess it's not one of the one's I used - or I'm keeping it in a zip for some
   reason??

I purchased these specific drives because there was some site that benched them
as being superior over all other similar drives when working in a RAID array. I'm
sure by now there are better one's available.

 

[ksz]

I used 5 or 6 different speed tests and everytime I find a new one I run it too.
I can search my HD and tell you the names if you're interested. Spotlight with
system files included doesn't turn up any apps with "AJA" or "Kona" in the name
so I guess it's not one of the one's I used - or I'm keeping it in a zip for some
reason??

Thanks for the answers. AJA KONA System Test is widely used to gauge read/write performance. It can be downloaded freely by clicking the previous link. Can you run the 256MB and 512MB tests and report the r/w performance?

 

[Tesselator]

Thanks for the answers. AJA KONA System Test is widely used to gauge read/write performance. It can be downloaded freely by clicking the previous link. Can you run the 256MB and 512MB tests and report the r/w performance?

Thanks for that. This is one I have seen and used before but I don't remember there being a video frame size selector (hmm...). I wonder what it's called on my HD? Hehe, anyway running the DL'ed one shows that I get nearly 500MB/s in some of the open spots so I'm not so sure about this test. I'm pretty sure that's not possible. The same test at 4BG with massive frame sizes shows 2 open spots that average about 380MB/s on the high side of those area's peaks. I included one at 16GB with a few very interesting hits in it. I'm trusting this software too much. I dunno if it would have anything to do with it or not but this partition is 96% full right now and I suspect highly fragmented. Anyway here ya go:


EDIT: Hehehe, with the system cache enabled it says I get 983MB/s.

EDIT: EDIT: I added two grabs like I mentioned in the with the cache on just for fun!

 

[Tesselator]

Hmmm, After a reboot testing what you asked (cache disabled) gives me:

Write: 240.5 MB/s
Read: 238.1 MB/s
on the 512 MB and:

Write: 236.2 MB/s
Read: 231.3 MB/s
on the 256 MB test.

This tells me that it's testing a lot more than the drive speed and therefore isn't a very useful tool for testing just drive speed alone. Also after the reboot with the cache enabled I get 1100 MB/s straight across the entire 16GB on the read and a higher number on the write average too.

EDIT: I dunno what's changed but several minutes after posting this I'm getting 1650 MB/s straight across the 16 GB with the cache enabled. Hmm.. You sure this software works?

 

[ksz]

Because it's the filesystem cache that's being turned on and off rather than the 16MB or 32MB drive-controller cache, the huge Read values are entirely believable -- this cache is stored in your computer's main memory.

The values you're seeing without the filesystem cache more accurately reflect the drive's raw performance. However, because drives have onboard cache (16MB or 32MB as mentioned), a second or third consecutive run of the same test will yield higher results than the first.

Hence, we should trust only the first result after a complete power down followed by power up (or after reading a whole lot of junk files to fill up the controller cache) because the drive-controller cache might not be purged simply by rebooting the Mac. It may require a complete power down.

The first set of numbers you posted, i.e. 130-140 MB/s, are the most representative of your RAID system's raw performance.

Thanks for taking the time to run these tests! Much appreciated.

If anyone can poke holes in my explanation, please do so.

 

[Tesselator]

I dunno about poking holes. That software seems rather poorly written to me.

Here it's saying that reading 256meg file (not in the cache) is well, I dunno how many kijila-gigs this is but:

 

[ksz]

It's reading Mathematica.app, not a 256MB file. Nevertheless, I would just stick to the read/write of an artificial file, which is the default option in the top pulldown.

Here's the RAID-0 performance for DAT Optic's eSATA PCIe card using AJA Kona: http://www.datoptic.com/cgi-bin/web.cgi?product=eSATA_PCIe8&detail=yes

Numerous other vendors and reviewers also cite results from AJA Kona. In addition, refer to this report from the Arizona Mac User's Group (AMUG) that compares 4 eSATA cards in RAID-0 with various quantities of hard drives, from 5 to 20. With 5 drives, the fastest card was able to achieve Read/Write performance of 239 MB/s and 216 MB/s respectively. Because the article was written in 2007, I would expect today's drives to yield better results, but not 2x better.

AMUG used DiskTester. According to their website:

DiskTester™ measures disk performance, especially RAID setups. But it can also be used for testing digital camera cards, network performance, or comparing single-drive performance with SATA, FireWire and USB. Its sophisticated testing approach offers the most consistent and reliable results available today.

Disktester is used for internal testing by several manufacturers of RAID and SATA hardware cards to test their products as well as performance-review sites (see below).

Because AJA Kona's numbers for "Disk Read/Write" are generally consistent with DiskTester, I am inclined to vindicate the AJA software. It may not be perfect, but it's perhaps close enough.

 

[Tesselator]

It may not be perfect, but it's perhaps close enough.

I'm going to stick with the ones that profile the disks and volumes all the way across the platter and display a nice curve. Even if all three of them are lying to me. Thinking that my RAID gives me 400MB/s in some areas and seeing just where that is somehow just makes me feel better. It allows me to make more intelligent decisions about where to place the partition boundaries too.