Best RAID config?

[Adam Gold]

(Apologies to those who have already seen this over at the Adobe User forums, but it's a little slow around there, so I thought I'd fish for opinions here as well...)

Many thanks to those who have given advice so far... but just wanted to be sure I'm doing the right thing.

The new workstation I've just ordered will have 7 x 1TB HDDs besides the system drive. I've been told putting them all together in a single RAID array (RAID 0, 3, 5 or 6) is the best and fastest.

But I'm still wondering if it might make sense to split out a drive or two for different media types and render destinations. While the video makes up about 99% of asset files, I'm wondering if we might use 5 disks for video, one for all other assets and one for previews and rendered files. I mean I know we could, but is there a point to that or am I just being silly and over-thinking this?

I'll be using Aspect HD with PPro CS3.

Thanks again to all who have helped with this whole system thing so far...

ag

[Allen Plowman]

raid 0 is for speed. if one drive in a raid 0 array fails, all the data in that raid array is lost.

[Adam Gold]

Right, but my question was whether anyone thought a single huge array is better than splitting out a few discs for other media types or destinations....

My main priority is speed.

[Chris Soucy]

It's an interesting question indeed.

I suppose it depends on just how your software works. If you can assign drive J as your input files and drive K for your output files for example, you can have both drive J & K doing reads/ writes simultaneously. If you can specify yet another drive for Audio (say) you can have that one operating independantly as well.


If this scenario is actually the case, it makes sense to set up three drives
J = 2 X 1TB; K = 2 X 1 TB; L = 3 X 1 TB (All Raid 0) or something similar.

I do not believe Raid 0 requires an even number of drives.

It also gives you the option of backing up extra special files to another drive in case you lose "the master", whichever one that is (although, given the size of this setup, it would probably be cheaper and more efficient to be using external drives for backup).


Alternatively, if your software can't work in this fashion, maybe just 2 drives;

J = 3 X 1 TB & K = 4 X 1 TB

Think you need to delve into the software for the appropriate info.


CS

[Robert M Wright]

It's difficult to conceive a circumstance where a single 7 drive RAID 0 array would offer any real world benefit (as opposed to some 2 drive RAID 0 arrays), and it would certainly increase the chances of a catastrophic failure dramatically.

[Adam Gold]

Thanks, guys. This has been my thought all along, as with Premiere you can set up 6 different scratch disks. But recently I've had people who know a lot more than I do recommend all drives in RAID 0, RAID 3, RAID 5 and RAID 6, as well as each media type to a different drive. Five variations, five opinions, one vote each. Annoyingly, all the Adobe docs tell you is what you CAN do, not what you SHOULD do in a a situation like this. (They do warn against using external and/or network drives, and make clear that simply partitioning a single physical drive does you no good at all, obviously.) I guess when they wrote the help files they didn't envision multi TB arrays.

Would five drives in RAID 5 give me the same speed and capacity as four in RAID 0, with the added benefit of safety? I'm willing to "waste" 1TB for security's sake, but not 2, and not if it sacrifices too much speed.

The kicker is that the project I'm working on now will generate about 3.5 TB of footage so I need at least one monster drive.

[Andrew J Morin]

Originally Posted by Adam Gold

Would five drives in RAID 5 give me the same speed and capacity as four in RAID 0, with the added benefit of safety? I'm willing to "waste" 1TB for security's sake, but not 2, and not if it sacrifices too much speed.

The kicker is that the project I'm working on now will generate about 3.5 TB of footage so I need at least one monster drive.

RAID 0 is courting trouble, IMO. RAID 5 is only "safe" if you have spare drives to sawp-in and rebuild the RAID. Having 7 drives in hand, you //could// designate 1 as a hot-spare, then a nominally 6 drive RAID-5 is probably your best option.

A more exotic solution that occurs to me:
2 or 3 drives in a RAID 0 for speedy working directories, and the remainder in RAID-5 with a hot spare for raw data and safe archives.

Of course, if your RAID *controller* craps out (has happened twice in my career) you're SOL either way. I like RAID-10 (creates a pair of stripe-sets and mirrors them, cuts storage volume in half but your data is in two places, and it's roughly as fast as a RADI 0; aka "RAID-0+1") but I understand that your need for space is topmost.

[John Sirb]

Adam to answer the other part of your question, I keep any audio and graphic files on a seperate drive (G) for graphics and (S) for Audio. Same physical drive jsut different partitions. and it worked fine

[Adam Gold]

Thanks again for all the responses. Again, Adobe is pretty clear that using separate partitions on the same physical drive does you no good at all; you need to assure that different types of files are on physically different platters with different read/write heads in order to enjoy a speed or efficiency boost. Having them on different partitions of the same physical drive is no different, speed-wise, than having them in the same directory on a single drive.

Any form of mirroring is out of the question. This whole issue of drive failure is, in my opinion, vastly overstated. I've had about 20 PCs containing about 30 Hard Drives over the past 20 years, with an average life span of about 5 years each, and have a total of one drive failure. I usually run out of space and move everything to a new drive before the old one has a chance to fail. I imagine I'll have an external eSATA petabyte drive hooked up before any of these have a chance to go legs up. And if I'm wrong I have the project file backed up on an external drive; in the event of a catastrophic drive failure then all I need to do is re-load the backup project file, highlight the clips and hit "Batch Capture" and walk away for a while. It'll kill a day but it won't destroy the whole project.

So my original question remains: Which is faster, 7 drives in RAID 5 or five in RAID 5 plus 2 JBODs, one for other input files and one for output?

I still think it would be the latter, but I can't get any evidence either way. If what Robert says is right about the diminishing returns with more drives in a single array -- and that makes sense to me -- then the latter config seems to me to be the best of all worlds. But I can't prove it.

Perhaps this just can't be answered and the only way to know is to try it out. At any rate, thanks for all who have provided input. More feedback is, of course, welcome.

[Mike McCarthy]

No question that a single array of all seven drives in RAID5 is the way to go. With seven spindles, Raid0 is way to risky for any long term work. A single array will always outperform dedicated disks, and Raid 5 adds security. If you need high performance, make sure you have a good Raid controller for maximum throughput. Integrated controllers are usually bad for Raid5, although they are great for Raid0 and Raid1. A dedicated PCI RAID card is worth the expense, for a couple hundred bucks. The drives cost more thhan that, and the Data cost WAY more than that. Regardless of whether you are editing DV or uncompressed HD, I would go with a single RAID5 array.

[Allen Plowman]

Originally Posted by Adam Gold

Right, but my question was whether anyone thought a single huge array is better than splitting out a few discs for other media types or destinations....

My main priority is speed.

are you willing to risk losing 100% of the data on all drives for the sake of optimum speed?
A single raid 0 is the greatest risk

[Don Blish]

Originally Posted by Mike McCarthy

No question that a single array of all seven drives in RAID5 is the way to go. With seven spindles, Raid0 is way to risky for any long term work. A single array will always outperform dedicated disks, and Raid 5 adds security. If you need high performance, make sure you have a good Raid controller for maximum throughput. Integrated controllers are usually bad for Raid5, although they are great for Raid0 and Raid1. A dedicated PCI RAID card is worth the expense, for a couple hundred bucks. The drives cost more thhan that, and the Data cost WAY more than that. Regardless of whether you are editing DV or uncompressed HD, I would go with a single RAID5 array.

I disagree with relying exclusively on RAID5 for video applications. I first built my machine with a 4 disc RAID5. It runs PremierePro2 + Aspect HD. Capture and editing were smooth as silk, but with long renderouts of .avi files or encodes of .mpg I could not reach 100% cpu utilization. [Task manager/performance tab] This is the sure way to find out if you have an input/output bottleneck. I added another pair of discs just for video work and configured these as RAID0. Now system and all irreplacable things are on the RAID5 and the current project is RAID0

With such huge drives now, I would configure four as "RAID10" This is effectively a mirrored pair of RAID0 drives. Fault tolerant and fast. [I'd buy a 5th disc and keep it on the shelf though]
[This is a built in capability of the Intel storage controller on the better 3rd party motherboards]

I was unsuprised with the RAID5 bottleneck because that was designed for commercial data processing with many random reads and fewer random writes. Since each write goes to two discs (the data to one and parity to another), adding more discs to a RAID5 does not really help the heavily pipelined, sequential nature of video work.

[before retiring I was a commercial DP department manager]

As for drives, buy good ones like Seagate, note the warranty period (5yrs?), and resolve to build a new system with new drives well before that.

[Ron Evans]

I RAID is fast for single file management but if one has a multi track/multicam/PiP project then a two disc array could be slower than just two discs with the files allocated to each disc. The reason is that in a RAID the files are split into pieces and allocated between the drives so that for a read operation both drives are having to seek more individual small files than is the case for separate drives that just read out a contiguous file at drive throughput speed( potentially 10 to 15 times faster than needed for DV or HDV). Modern drives have improved the throughput but seek times have not changed much in the last few years so many seeks will slow down the system. For temp files , in the case of a single RAID system with all on this RAID then the RAID will have to read multiple files and also write temp files at the same time too. Not the ideal for speed. My preference is OS drive, temp drive and separate drives for storage. Most projects are multicam DV so I don't see the need for RAID at all, the processor is the bottleneck. But you do need to manage where the files go!!! IF you don't want to manage where the files go then a couple of RAID 5 would be called for with the temp/scratch a RAID 0 as this may be transient most of the time and can always be re rendered anyway.

Ron Evans

[Allen Plowman]

I store my work on a raid 1, but I move my work in process to a raptor 10,000 rpm drive. I prefer that to a raid 0. there is a time factor to move files, but for my arrangement, it suits me.

[Adam Gold]

This is great -- just the kind of debate I was hoping for. I'm not really any closer to a solution as we now have about ten solutions with one vote each, but I'm very appreciative of the time everyone has put into thinking this through.

So for those who aren't totally bored with this issue -- and I don't blame you if you are -- I think it wold help if we split this into two parts. The first part is what RAID type -- 0, 3, 5, 6 or a billion -- we should use. I think we can shelve that for the moment as that really isn't the hard part of the equation.

The hard part is how you group the drives regardless of your RAID type.

Obviously your video has to be in a single directory, or else it gets way too complicated, so it would be on one group of disks.

Then you have your audio. If the audio is on the same array as the video there's no way to avoid the possibility of having to retrieve both audio and video simultaneously from the same platter at a given moment. But by putting audio on a separate spindle you can assure you'd never have to access both at once with the same read head.

And then there's writing (rendering) and the only way to assure you'd never have to write to the same disk that you're reading from is to again use a separate physical drive.

If all this was together in a single large array I can't see how there wouldn't be a potential bottleneck.

I guess that's the crux of my question -- is my thinking here wrong? I can't see any way a seven disk array, regardless of RAID type, could be as fast as 5 + 1 + 1 in the way I described in my first post. I'm just not willing to spend every dime I made from my last gig just to see eight cores crawling along at 35% utilization. I need this baby to scream. I need to feed the data to the CPUs as fast as they can handle it.

God, I just wish I could set up parallel systems side by side and try it a bunch of ways. Again, we'll leave whether it's RAID 0, 1, 2, 3, 4, 5, 6, or 10 to another time.

Thanks, everyone, for bearing with me on this.