What's important for a RAID?

[sechs]

I'm getting ready to put together a RAID 0+1 array and have a couple drives in mind as possibilities for it; however, I'm really sure for what I should be looking.

What kind of performance characteristics are good (or best) for this kind of array? Which ones are more important than others?

For example, is seek time important, and would it be worth trading 0.5ms for better throughput?


Thanks in advance!

 

[DaveSimmons]

You might get more responses if you edit your post to include what you want to use the RAID for. And RAID 0 might not be necessary (leaving you with just RAID 1) given that single-drive throughput has climbed steadily with the increase in platter density.

If money is no object and high speed or lower failure rate is critical, SCSI RAID with 15K drives.
If money is no object and storage capacity is critical, perhaps an array of Maxtor 250 GB IDE drives.

 

[AluminumStudios]

I don't think .5 ms access time will make as much of a performance difference as your stripe size.

Many RAID cards tend to have a default 64 kb stripe, but if you work with large files then that stripe size can be a performance limiter. I work with video and use a 256 kb stripe on my 2 WD40bb drives in RAID 0. Any smaller and I would suffer a performance hit for my application.

If on the other hand you work with smaller files, then a smaller strip might give you a speed boost.

So my answer is dont' sweat the little tech details. But decent drives in your price range, then configure the RAID in a way that makes sense for your usage patterns.

 

[sechs]

RAID 0 lending some performance over JBOD and mirroring making me feel better, 0+1 seems like the right choice. That's water under the bridge anyway, a decision already made. I've bought a Highpoint 1540 card for the purpose, so I have a choice of stripe sizes; that's a matter which will be addressed later.

Is it that nobody knows what kind of performance characteristics are important in a RAID, or that it largely doesn't matter?

 

[Kinesis]

Anyone want to suggest to my why the average programmer / gamer would go with a RAID array and which type? Better yet, where is a good place to get info on RAID arrays and their applications ?

 

[sechs]

Don't hijack my thread.

 

[JKing76]

Originally posted by: Kinesis
Anyone want to suggest to my why the average programmer / gamer would go with a RAID array and which type? Better yet, where is a good place to get info on RAID arrays and their applications ?

An average programmer might like RAID1 (mirroring), which basically provides an automatic backup of all his source code. An average gamer might like RAID0 (striping), which increases transfer speed -- although this would only be an issue if he's playing a game that's swapping a lot between memory and harddrive.

 

[sechs]

But what about the drives?

 

[Kinesis]

Sorry Sechs, I thought the question was answered by AluminumStudios .... May deepest apologies.

:-(

 

[JKing76]

Don't hijack the new topic.

 

[AluminumStudios]

I've worked with IDE RAID at home and SCSI RAID at work and would tend to personally recommend:

programmer: RAID 0+1 because RAID 0 gives you a speed boost and I know that compilers really hit the hard drive in the creation and reading of many small files. Also for a programmer who probably generates and reads many small files a smaller stripe size like 64 kb would probably be beneficial since that would allow any file over 64 kb to benefit from being written/read to both drives in parallel. Programmers I assume also want reliablity so 0+1 provies redundancy in the event of a hardware failure.

gammer: RAID 0. This will give you a speed improvement without costing much. You can spend the extra $$ on a video card or what not. I'm going to guess that a hard drive failure for a gamer won't be as catastrophic since they can just reload the games from the CDs and any saves, etc. will be small enough to back up on CD another hard drive.

So I guess if you are a programmer+gamer then RAID 0+1!

I'm a video editor myself so RAID 0 definitly helps me. I actually work with files on my RAID, then back them up manually to a separate hard drive that I have. RAID 1 protects against hardware failure, but not human failure ... so if I mess up my project (as I have) RAID 0 won't let me step back in time whereas keeping duplicate copies on another hard drive will.

 

[sechs]

Now that everyone has discussed RAID, let's talk about harddrives!

 

[Jeff7]

Originally posted by: sechs
Now that everyone has discussed RAID, let's talk about harddrives!

Don't hijack your own thread!
j/k

What about them specifically do you want to know? 7200rpm is better than 5400rpm. Western Digital's Special Edition drives with the 8MB buffer seem to be popular, and are some of the fastest IDE drives out there now.

 

[Bovinicus]

Make sure all of your drivers in the RAID array are the same, or at least similar. Slower driver will hold RAID arrays back when it comes to performance. As well, smaller driver will make larger drives only use the amount of space equal to the smaller driver. 0+1 is the way to go if you can afford 4 drives. Data reliability + Performance.

 

[sechs]

Originally posted by: Jeff7What about them specifically do you want to know? 7200rpm is better than 5400rpm. Western Digital's Special Edition drives with the 8MB buffer seem to be popular, and are some of the fastest IDE drives out there now.

What performance characteristics are important?

In discussion elsewhere, I have determined that spindle speed (a physical characteristic) is irrelevant, and cache size (also a physical characteristic) is not relevant (in fact, more may make your RAID slower!).

 

[Mark R]

The characteristics are the same as for chosing a single hard drive, and depend primarily on the pattern of use - e.g will you be manipulating huge video streams, or accessing huge databases in a random fashion?

There are however, a few minor changes:

If you are using RAID 0:
then sustained transfer rate is less of an issue than for a single drive - as with 2 drives transferring simultaneuosly you may hit a bottleneck elsewhere.
Because RAID 0 is dependent of both drives reading information, access time is important (essentially the array takes on the access time of the slowest drive). Because transfer rate is higher, seeking may become the bottleneck (depends on your type of use - databases particularly).
In fact, with RAID 0, unless you spindle sync the drives (a SCSI only feature), you actually degrade the access time of the array further by, on average, a time equivalent to half a revolution. Spindle speed is therefore very important in a RAID 0 configuration if your application is likely to be seek limited (Not only does a faster spindle speed decrease the average access time, but it decreases the access time penalty caused by running unsynchronized drives).

With RAID 1:
Data transfer rate is equal to that of one drive, and access time is roughly 1/2 that of a single drive (if your controller distributes read requests intelligently, some don't, and only read from one drive). Transfer rate therefore becomes relatively more important than access time (and spindle speed) in this configuration.

 

[sechs]

If I hadn't mentioned it earlier, the RAID will be IDE.

Your comments and explanations make a lot of sense, Mark R. The thing is, I'm putting together a RAID 0+1.

The two drive models at which I am looking most seriously differ in that one has better seek and the other better throughput. On that basis, which one would be better?

 

[sechs]

After more digging, I found this little tidbit of information. (It's good to know that I've been looking at the most important characteristics.)

Since I pretty much plan on putting everything on array, I presume that I'll be doing a lot of random accessing and, therefore, need to emphasize seek. Does this sound correct?