[cube-dude]

Next weekend I'll become a first-timer to RAID and, while posts like this are insightful, what I really could use is some help with the software setup process itself. I came across a Mac|Life article (for those of us who subscribe, pages 91-93 of the February 2007 issue) and would appreciate your critique of the instructions and any other expertise:

http://www.maclife.com/article/rock_...n_your_mac_pro

My desire is speed and redundancy, equally, with a 2 x 2.8 Quad that will arrive any day now. Just like in the article, the Mac Pro will have 4 x 500 GB drives and no RAID card. Unlike the article, obviously the OS will be Leopard, not Tiger. (Relative to the step by step instructions, any Disk Utility differences I should be aware of?)

So, overall, what do you think of these instructions, and what (if anything) would you do differently or caution me on? Being a RAID newb, go ahead and feel free to talk to me like a fourth grader.

Some personal examples on choosing block size would be great toward helping me decide Step 6. The article's examples here (audio/video or databases) seem rather "either/or" to me, when in reality this multi-purpose Mac will be handling both large and small files alike. Thoughts?

Additionally, can someone please better articulate the very last sentence in Step 9? If one drive in this setup of four were to fail, are they saying I can simply swap the bad drive for a new 500 GB and automatically, painlessly go about my business, or would the RAID have to be completely reconfigured? Anyone experienced this?

Finally, since this will be a new, never-booted Mac right out of the box, do I immediately jump right into setting up the RAID with the System Install disk (Step 4), or should I set up a basic admin account first before rebooting from the Install disk? What's best?

Lots and lots of questions, so I hope they make sense! Thanks all.

I'll be traveling without access over the next few days (argh, of *course* I'm away when the Mac Pro arrives), so I'll check in mid-week when I can. Really looking forward to my inaugural RAID.

[mduell]

First, I'd really consider buying the RAID card. The difference between software and hardware RAID is like night and day.

The only place I disagree with those directions is that I'd do a stripe of mirrors instead of a mirror of stripes. Cuts recovery time in half.

For block size, take the default.

If one drive fails, you'll still be able to access your data like nothing happened. Replace it (be sure to yank out the right one!) and go find a repair button in disk utility.

There's no point in doing anything before you create the RAID array, since creating the RAID array will wipe out everything.

[SierraDragon]

Originally Posted by cube-dude 

My desire is speed and redundancy, equally... ...the Mac Pro will have 4 x 500 GB drives

Although "speed and redundancy" are admirable desires for all of us, RAID arrays take disks to configure the arrays, and four 500 GB drives is not really very much total mass storage to be configuring into multiple RAID arrays when mirrored arrays are part of the configuration.

Important IMO before defining RAID is for you to determine what apps will be run and how both on site and off site backup will be done. Without that info configuring RAID is premature. Also you must first decide if you intend to use Time Machine, a huge mass storage hog.

Note that hard drives start slowing at something like 50% full and can get very slow or even unstable above 85% full; a good rule of thumb is not to exceed 70% full. 2x500 drives in a RAID 0 array therefore provide ~700 available GB while 2x500 drives in a RAID1 array provide ~350 available GB. RAID0 is ~2x as fast and 2x as likely to fail as is a single drive, RAID1 is real-time redundant.

The 350 GB available from 2x500 RAID1 generally is not a good backup capacity for the 700 GB available from a 2x500 RAID0 array.

IMO for most users the real-time redundancy of RAID1 (mirror) is not appropriate. Automated hourly or daily backup to non-RAID drives is simpler and less expensive, especially when working within a limit of only 4 drives total. Generally only real-time financial transactions and/or large drive arrays justify RAID1.

I am happy to suggest drives configuration, but on and off site backup plans are required pre-configuration info. The (70%) functional capacity of every array and non-array drive must be written down before configuring the RAID setup. E.g. a stripe of mirrors instead of a mirror of stripes affects capacity and may not implement well without a hardware or software RAID controller that specifically allows RAID 10. In addition to very high cost per GB a RAID 10 configuration tends to not be very scalable as future needs evolve.

Wiki has a good required-reading RAID white paper at
Standard RAID levels - Wikipedia, the free encyclopedia

-Allen Wicks

[mduell]

I thought it was safe to assume the OP has a separate backup strategy with external drives. Working volume and backup are completely different issues.

The OP said nothing about backing up 700GB onto a 350GB array. He's looking at creating a 1TB RAID10 array for his working volume.

A stripe of mirrors instead of a mirror of stripes has no impact on capacity, with or without your 70% rule.

In the last few months I've become a huge fan of RAID1 for three very different applications: servers, workstations, and home media boxes. RAID10 beats RAID5 in terms of recovery time from a drive failure (zero), performance (particularly on writes) and cost (good, and even bad, RAID5 controllers are expensive). RAID0 suffers from the same recovery time problems (it doesn't take many hours of array rebuilding and data copying to pay for a pair of 500GB or 1TB drives) as RAID5.

[OreoCookie]

Sierra Dragon is right on the money.
A RAID0 will give you speed, but no redundancy. A RAID1 will give you redundancy, but no speed (advantage). A RAID1 is not a substitute for a backup. I think it'd be easiest if you tell us what you want to do with your computer.

[SierraDragon]

Originally Posted by mduell 

A stripe of mirrors instead of a mirror of stripes has no impact on capacity, with or without your 70% rule.

You are correct, sorry Mark. Either of those approaches (RAID10 or RAID0+1) takes 2TB of gross capacity before overhead and makes it into 1TB of gross capacity before overhead, or ~700GB appropriately usable mass storage and with no provision for a separate scratch disk.

IMO for 4-drive single workstation setups like this one "recovery time" is moot. Recovery, if ever, will only be needed on average every few years. Absolute throughput is more important.

-Allen Wicks

[SierraDragon]

I may have implied a dislike for RAID 10 above. That is not the case. I think RAID 10 is excellent, ideal even, for setups that need real-time redundancy plus speed. I just think that few single-workstation setups outside the financial world need real time redundancy, and those that do justify more than 4 disks total in the full setup.

I too am brainstorming a new 8-core 2.8 GHz MP mass storage setup and at this point I am thinking:

• Apple drive A for system/apps, 500 or 750 GB.
• Third party drives B&C as RAID0 array, each drive 1TB or larger.
• Third party drive D for on-site backup, sized 1.5TB or larger.
• Third party external drive E for off site backup 1.5 TB or larger. This drive partitioned with one partition of Drive A size for Super-Duper cloning of Drive A; a second partition for backup of the B/C array.

I am still looking for more real world performance info, and with more info instead may add another external drive for on site backup and make B&C&D all part of the RAID0 array, adding 50% more capacity and substantial additional speed to the RAID0 array.

Drive A and RAID0 array drives will for speed not be allowed to exceed 50% full. Backup drives will for stability not be allowed to exceed 85% full. IMO it is important to plan in advance how full drives will actually be allowed to become and then stick to that plan.

I am still debating among Apple's RAID card, Apple's RAID software and SoftRAID <http://www.softraid.com/>. Hardware RAID is strongest but cost has me currently leaning toward SoftRAID. Note that SoftRAID does not require identical volumes which adds a lot of flexibility in the long term. I talked to other consultants and spent time at the Expo booth and was favorably impressed.

I do believe that OS 10.5.2 may impact what is optimum for me due to the likely changing relationship between Aperture's Vault backup and Time Machine. I am ignorant regarding Time Machine; if it is wonderful it may mean one more drive.

Comments and crtique are much appreciated.

TIA

-Allen Wicks

[mduell]

Originally Posted by SierraDragon 

I may have implied a dislike for RAID 10 above. That is not the case. I think RAID 10 is excellent, ideal even, for setups that need real-time redundancy plus speed. I just think that few single-workstation setups outside the financial world need real time redundancy, and those that do justify more than 4 disks total in the full setup.

I too am brainstorming a new 8-core 2.8 GHz MP mass storage setup and at this point I am thinking:

• Apple drive A for system/apps, 500 or 750 GB.

Bad idea. Take the 320GB from Apple and buy a 500/750/1000GB for the price they want for the 500/750/1000GB upgrade. Free 320GB drive.

Originally Posted by SierraDragon 

I am still debating among Apple's RAID card, Apple's RAID software and SoftRAID SoftRAID Home. Hardware RAID is strongest but cost has me currently leaning toward SoftRAID. Note that SoftRAID does not require identical volumes which adds a lot of flexibility in the long term.

Software RAID may be able to catch up in bandwidth in some benchmarks, but it can't compete for I/Os. Also, hardware RAID supports arrays with varying disk sizes and (if you have good software to manage the RAID with) allows you to grow the array once all disks are a larger size. It also generally has less shenanigans with weird things happening and corrupting your array in my experience.

[cube-dude]

Hi Mike, Allen, and OreoCookie. The RAID newb just returned from travel and is grateful for everyone's guidance AND my new Mac Pro, which arrived Wednesday. Thanks.

While I'd definitely prefer letting a RAID card do its thing, the budget won't allow for more hardware this time around. And, yes, I am operating under the premise of working volume redundancy and external backup as two separate strategies. I am trying to get the terminology and technique down, so please correct me if I'm wrong: the RAID in the article is a mirror of stripes (RAID 1+0), not a stripe of mirrors (RAID 10), correct?

Thanks to your comments and after reading up more on RAID pros and cons in general, I feel that a stripe of mirrors netting 1TB is going to work best for me. As a first timer to RAID setup, what steps in the article (5? 7? 8?) exactly do I need to do differently to achieve a stripe of mirrors instead of a mirror of stripes? (Remember, fourth grader talk welcomed!) TIA

Maybe the above will become obvious once I'm actually in Disk Utility, but I'd much prefer to have detailed instructions in front of me first since I really don't want to screw up this weekend.

[mduell]

RAID 0+1 (RAID01) is a mirror of stripes. RAID 1+0 (RAID10) is a stripe of mirrors.

For following those steps, create two mirrors first (Ham and Cheese), then create the stripe (Sandwich) from those arrays.

[cube-dude]

Originally Posted by mduell 

For following those steps, create two mirrors first (Ham and Cheese), then create the stripe (Sandwich) from those arrays.

Got it. Did it. Then, a few seconds after clicking Create to make Sandwich, a Disk Utility dialogue appeared and said it could not continue. I was prompted to shut DU before Sandwich completed. Argh.

Reopened DU. Put the Ham and Cheese mirrors (which I noticed this time no longer appeared at the end of the four drives in the sidebar list, but closer to the top) into the Sandwich stripe again, clicked Create, and I now have a striped set of ~1TB as anticipated. However, underneath Sandwich (Online, in green) there is Ham (with now only one drive underneath it, disk1s2) and Cheese (also now with only one drive underneath it, disk3s2).

Ham and Cheese both appear as Degraded, in yellow. Shouldn't there be one more drive underneath each?

Additionally, the four disks (appearing in red) in my DU sidebar show as (in order): disk0s2, RAID Slice, disk2s2, RAID Slice.

Looks like I will be redoing my RAID over again from the start, and wanted to keep you all in the loop. TIA

[cube-dude]

Originally Posted by cube-dude 

Then, a few seconds after clicking Create to make Sandwich, a Disk Utility dialogue appeared and said it could not continue. I was prompted to shut DU before Sandwich completed.

Well, it happened again. The exact error: "Disk Utility has lost its connection with the Disk Management Tool and cannot continue. Please quit and relaunch Disk Utility."

Click OK, relaunch, and now the order in the DU sidebar is: disk0s2, RAID Slice, RAID Slice, RAID Slice, Ham (in red), Cheese. I am stumped and look to greater wisdom.

[mduell]

This is one of the ways software RAID sucks. It's just a bunch of flaky hacks all around.

Consider the value of the billable time you've already wasted on this and buy the RAID card.

[tadd]

A Mirror COULD be faster during reads than a single drive. It would require the RAID software or hardware to perform the reads in parallel (easy with SATA) getting different parts of the files being read from each of the two drives, giving 90MB/sec read data rate (for example) where reading from a single drive would give 45MB/sec read data rate (from my example). Writes to a mirror are at the same rate as writes to a single drive, i.e. 45MB/sec per channel to each drive vs 45MB/sec on a single channel to a single drive.

I recommend playing with the RAID software just to find out how cool or rotten it is but my personal PowerMac G5 has no RAID. I have four separate SATA drives and a 2TB USB2 connected external box (Western Digital 2TB My Book Pro). The 2TB drive is my Time Machine backup drive.

BTW I worked for the RAID division of Adaptec for 5 years ending in 2005. Alas they started killing off Mac support just as I was starting to work there.
Tadd

[SierraDragon]

Originally Posted by tadd 

...personal PowerMac G5 has no RAID. I have four separate SATA drives and a 2TB USB2 connected external box (Western Digital 2TB My Book Pro).

Tadd-

2008 MPs loaded with RAM and 8800GT graphics have far superior performance to the best G5, making mass storage a significant bottleneck for those of us handling images professionally. Ergo the need for RAID0 arrays.

-Allen Wicks

[cube-dude]

Originally Posted by mduell 

It's just a bunch of flaky hacks all around. Consider the value of the billable time you've already wasted on this and buy the RAID card.

Well, this just sucks (my scenario, not your response, which I cannot dispute).

More than the time spent and dollars involved, what irks me the most is that 1) Apple's own RAID software is crappy enough to encourage the sale of Apple's own RAID hardware (how convenient), and 2) were said software up to snuff, my abilities have grown to the point, thanks to reading and recommendations, where I could configure RAID in a matter of minutes without instructions in front of me, something I could never have done just a few weeks ago.

Que sera, sera, but I do want to thank you for your guidance.