Raid 0, safe enough?

powadha · 08-08-2005, 08:45 AM

I'm thinking of using Raid 0 in my new pc with two 200 gig hdd's. I've read though that using raid 0 can be unsafe (an error on 1 disk makes all the data useless) I'm wondering if Raid 0 makes a disk error occur sooner then when using just 1 disk and no Raid.

Regards

jtshaw · 08-08-2005, 09:03 AM

Raid 0 will increase your disk I/O performance but will provide no redundancy so it isn't an effective method of insuring the safety of your data.

Basically, if you use RAID 0 you just have to understand that the raid array isn't providing you with any kind of backup of your data and thus you have to know to back your data in another way.

powadha · 08-08-2005, 09:21 AM

That's indeed the downside to this setup. The hunt for speed makes it sound very interesting but loosing data on all partitions at onces (since all data is split to the two disks) in case of a drive failure makes me go toward a normal setup.
Perhaps a third drive outside the array used for backing up the most important data on regular intervals could make it a bit safer.

I'll guess I have to deside how much I need the HDD speed compared to how important my data is.

Thanx

jtshaw · 08-08-2005, 09:25 AM

If you are willing to explore a 3 disk option why not look at RAID 5? It provides stripping and a parity disk so that the loss of any one disk in the system doesn't cause a lose of data.

RAID 5 doesn't do writes as fast as a simple strip array (it has to calculate the parity bit) but it is very fast for reading and provides redundancy.

powadha · 08-08-2005, 09:38 AM

That is the solution I guess! Downside is that I was planning on buying two and not three new disks. I read that (at best) the disks should be of equal size and speed. Since the new pc will have sata drives it would be a wast to connect the older WD non sata drive in the array, since it will slow down the whole concept. Well, have to do some more saving up I guess.

Thanx for the info!

IsaacKuo · 08-08-2005, 10:04 AM

Another option to consider is partitioning each disc into a small partion and a large partition. The small partitions can be combined with RAID0 for speed, with regular backups.

There's a question of what you need the speed for, and also a question of whether you'll actually get any speedup at all. RAID0 is good for speeding up loading/writing large files, but it doesn't help seek times (it can even make seeks SLOWER).

For many purposes, what you really want is to improve seek times. For that, you want RAID1--and your data will be stored in a safe, redundant manner also!

IsaacKuo · 08-08-2005, 10:13 AM

How big is your old WD drive? It may be as fast as the SATA drives. Even if not, you can RAID the fast drives and use the old WD drive for regular backups of the more important files (assuming it's not big enough to hold them all).

Another, more obscure possibility, is RAID4 with the old WD drive as parity. Write speed will be limited by the WD drive, but read speed will be the same as RAID0 on the SATA drives.

Note that with Linux's software RAID, you can combine individual partitions rather than just whole drives.

powadha · 08-09-2005, 08:38 AM

Didn't know the partition option! That creates a whole new way of setting up the array....The old drive is a 120 gig WD 7200 drive (8 mb) I guess I have more reading up to do on Raid and the Raid partition options.

Thanx for the info!

Pete M · 08-09-2005, 01:12 PM

powadha

Heres a comparison of the performance hit between RAID0 and RAID1

Both machines have similar hardware and are using pure software raid each with 2 x Western Digital Raptors

Only one partition on each disc is a member of the raid array

RAID1

Code:

# hdparm -tT /dev/md0

/dev/md0:
 Timing cached reads:   2920 MB in  2.00 seconds = 1459.49 MB/sec
 Timing buffered disk reads:   94 MB in  3.01 seconds =  31.27 MB/sec

RAID0

Code:

#  hdparm -tT /dev/md0

/dev/md0:
 Timing cached reads:   3804 MB in  2.00 seconds = 1902.29 MB/sec
 Timing buffered disk reads:  376 MB in  3.01 seconds = 124.73 MB/sec

Pete

IsaacKuo · 08-09-2005, 02:31 PM

From what I've read about Linux RAID, hdparm is NOT able to accurately measure performance.

Regardless, even assuming the numbers are accurate read speed doesn't tell the whole story. For many things, seek speed is far more important.

When you're browsing digital photos which are between ~1 meg in size, the read time difference between .03 seconds and .01 seconds is small compared to the seek time of .1+ seconds. RAID1 can significantly reduce seek times, depending on the situation, while RAID0 can slightly increase seek times.

Also RAID1 has overall read rates as good as RAID0 when multiple files are being accessed at once.

Crito · 08-09-2005, 02:50 PM

Quote:

Originally posted by powadha
I'm wondering if Raid 0 makes a disk error occur sooner then when using just 1 disk and no Raid.

The odds are exactly the same. In fact, a single HDD is itself actually comprised of mulitple platters and heads which spread the I/O very similarly to RAID 0. In any case, it doesn't matter if you flip a coin once or twice, the odds of it coming up heads or tails is still 50/50.

Electro · 08-09-2005, 04:04 PM

It is safer to use RAID 10 (stripping and mirroring). Software RAID 5 can have some problems. The utility hdparm only does raw performance. The data throughput of your setup will be around 30 MB per second to 60 MB per second because of filesystem, controller, hard drive, and bus over head.

Crito · 08-09-2005, 04:42 PM

... but you'd need four drives for RAID 0+1 (aka RAID 10).

Electro · 08-10-2005, 02:14 AM

Quote:

Originally posted by Crito
... but you'd need four drives for RAID 0+1 (aka RAID 10).

Yes, but you still get performance if one drive goes. With software RAID 5 it slows down to a crawl when a hard drive fails and in some services the system can freeze. If powadha wants to stick with minimal hard drives as possible, powadha could use a 400 GB hard drive to store the journal while two 200 GB hard drives are setup as RAID 0. This setup will seem that is a RAID 4, but the write performance will be increase by 10% to 15% with out any overhead because the desire filesystem (EXT3, ReiserFS, Reiser4, XFS, JFS) does it automatically.

Crito · 08-10-2005, 11:15 AM

Yeah, I recommend a hardware-based RAID controller for RAID 5 too, if for no other reason than to off-load the CPU. I guess he needs to weigh the cost of a RAID controller vs. the cost of a fourth drive. Used to be the controller was cheaper (about $1/meg for a HDD in 1992), but nowadays disk drives are dirt cheap (less than $1/gig.)