How would YOU set up your disks?

[Adam Gold]

I asked about this a year ago when I was first setting up my system, but there wasn't any clear consensus then and there have been a few threads touching on this since, so it seems there are more than a few people interested in this.

In my workstation I've got 7 x 1TB internal SATA drives (not counting the system drive, which is not even on the RAID controller). When first setting them up I couldn't decide whether the best performance could be achieved via a huge RAID, separate disks with one type of media on each, or some hybrid. I elected to go with one huge RAID5 with a hot spare, netting me about 5TB capacity, but I still keep wondering if assigning each media type to a separate drive might be faster as we could then be sure we're not trying to read and write to the same drive at any given time, nor would we ever be trying to get multiple media reads from the same disk at once.

Using CS3 with Cineform Aspect, but have ordered upgrades to both. Pretty tricked out system but I want to keep that part out of it and just focus on the disk setup, all else being equal. Have been pretty happy with disk throughput so far but am always wondering about that perfect balance of performance and stability. My intention in the next few weeks is to quadruple-boot system with XP 32, XP64, Vista 64 and Win7 and try to do controlled experiments with each, so I want the OS to be the only variable.

So you're staring at 7 drives that could be configured any way you like. What would YOU do?

[Mike McCarthy]

One large array is the way to go in 99% of cases. Not much of a conversation to be had there, but the big questionthat should be asked is: how big of a stripe size to use? I prefer larger stripes, but certain controller documents recommend large stripes for database and such, and small stripes for video and media files. That always seemed backwards to me, and I have good results with large stripe sizes.

Obviously having a hop spare cuts your performance and capacity by 15% or so. You have to decide if that drop is worth the margin protection of an automatic rebuild.

[Adam Gold]

Originally Posted by Mike McCarthy

Obviously having a hop spare cuts your performance and capacity by 15% or so. You have to decide if that drop is worth the margin protection of an automatic rebuild.

Over the past year, my rig has been in the shop more than it's been used. We have had repeated RAID failures (always on the same channel no matter how many times the actual disk has been replaced) and they have just now admitted that there was a faulty connector on the backplane, and it's finally been replaced. It took them a year and six trips into the shop before they discovered this, even though I asked them to check this after the second failure. But I digress.

That hot spare saved my bacon a time or two. If I can get it running flawlessly for more than a month, I'll kill the hot spare and make it part of the array.

The theory behind one huge array being best is that you have like a zillion read/write heads distributing the loads at all times, so faster performance? That was my thought based on your recommendation a year ago. But I still wonder if the read performance (crucial in rendering and output) would be better if there were separate drives assigned to each type of asset, and to different drives for output, so we could be sure we weren't pulling from the same part of the same drive for multiple streams. I guess I could see it both ways, and in the end the differences may not be all that major.

[Roger Wilson]

Here is how I setup my system:

2 drives mirrored - This is the system drive and where programs are installed
2 drives stripped - This is the drive I do all of my work on, this allows for both fast read and write.
3 drive in RAID 5 - This is where I save/archive projects to. RAID 5 provides a good balance of space and redundancy.
1 drive - Used for Windows pagefile, and Temp files

Hope this helps,
Roger

[Adam Gold]

Interesting strategy. So you have your projects and all your assets on the striped drive?

What's your feeling on RAID 5 vs 3?

[Tripp Woelfel]

From a guy who used to work for several RAID manufacturers, albeit ending five years ago so some info might be dated, here some thoughts.

More disks are always faster. Big RAIDs are good for performance.

You could convert the hot spare to an active drive for more performance. Hot spares are essential to unattended operations, servers and the like. Unless you do a lot over overnight and weekend renders, a cold spare might be all you need. The right decision for you depends upon your security comfort level.

Large stripe size recommendations for databases seems very odd. If you modify part of a stripe the array needs to write the whole stripe. Bigger takes more time. Since video is read, not write intensive, large stripes are the ticket.

One company I worked for made only RAID 0, 1 and 5 and also made video servers. They never used RAID 5 for video but that was many years ago. Traditionally, RAID 3 has been preferred for video but things might have changed. Not sure this matters as much as stripe size, but I've already had my evening dram.

[Peter Manojlovic]

I've got a system drive..(C)
Data Drive for project files and other assets (D)
Rendered effect drive (Preview drive) (E)..
Capture drive (F)...Your choice of Raid configuration..

At any point, any of these drives can get accessed for playback or render, so it would be less taxing on your system to have the playback come from several physical drives for each function..


Good luck!!!

[Kevin Amundson]

I have:
(C) A single 140GB HDD that has the OS & Programs on it.
(G) Four 232GB HDDs configured with RAID 5 for a total of 695GB that stores all my media(video, audio, images, ect.).
(D) A single 300GB HDD for render/previews.

I almost exclusively edit HDV.

[Jon Shohet]

Could you please elaborate on the stripe size?
- What size would you recommend for a raid-0 / raid-5 media array?
- And does it make a difference if you also put your scratch files and/or renders on the media array (since you say large stripe size works well with read-intensive oriented data) ?

thanks :)

[Roger Wilson]

Originally Posted by Adam Gold

Interesting strategy. So you have your projects and all your assets on the striped drive?

When working on a project I have all the assets on the striped drive (RAID 0). This gives the best performance, but has no redundancy (in case of drive failure). All of the assets are checked in to a digital asset management system on a server. When ever I want to create a snapshot of what I'm working on, I just check the items in.

What's your feeling on RAID 5 vs 3?

From Wikipedia(1):
RAID 3 is a striped set with dedicated parity or bit interleaved parity or byte level parity. This mechanism provides an improved performance and fault tolerance similar to RAID 5, but with a dedicated parity disk rather than rotated parity stripes. The single parity disk is a bottle-neck for writing since every write requires updating the parity data. One minor benefit is the dedicated parity disk allows the parity drive to fail and operation will continue without parity or performance penalty.

RAID 5 is a striped set with distributed parity or interleave parity. Distributed parity requires all drives but one to be present to operate; drive failure requires replacement, but the array is not destroyed by a single drive failure. Upon drive failure, any subsequent reads can be calculated from the distributed parity such that the drive failure is masked from the end user. The array will have data loss in the event of a second drive failure and is vulnerable until the data that was on the failed drive is rebuilt onto a replacement drive.

I've never used RAID 3. I use RAID 0 for performance then have my archives on RAID 5.

(1) RAID - Wikipedia, the free encyclopedia

[Tripp Woelfel]

Originally Posted by Jon Shohet

Could you please elaborate on the stripe size?

Stripe size is the amount of data written to the data drives in the array in a single "pass". For example, in a five drive RAID 3 (four data drives and one for parity) with a stripe size of 40MB, the array will write 10MB to each of the four data drives and then compute the parity to write to the fifth. (Numbers for example purposes only. Not a recommendation). So...
Stripe size=Amount of data written to each data disk * number of data disks.

This is important. Think of a stripe like a block of data on a single disk. When you write a file, even a tiny one to disk, it writes a whole block. When you write the same file to an array, it writes a whole stripe. So in the previous example, writing a tiny file on disk would consume 40MB of space on the array.

Originally Posted by Jon Shohet

- What size would you recommend for a raid-0 / raid-5 media array?

I wouldn't. It's quite a black art. If you feel comfortable setting your own block size when you format a disk then you're up to the task of making a good guess on stripe sizes.

Originally Posted by Jon Shohet

- And does it make a difference if you also put your scratch files and/or renders on the media array (since you say large stripe size works well with read-intensive oriented data) ?

I misspoke. Large stripes are better for large file sizes, like video. Smaller stripes better for databases with small reads and writes. Consider what I said in the stripe definition above.

And no. Don't put your swap file on the array. The risk/reward ratio isn't good.