Will RAID 0 help on a small network?

The customer has 10 XP workstations and one XP computer that is acting as their "server", running a standard IDE har drive.  They now want to add workstations on the network, and want to go to Win 2003 SBS on a real server.  They are running Accpac accounting, most of which runs off the server (the programs, data, and the database engine to some degree), and they also want to start using Exchange server.

I think I understand the benefits of Raid 1 for mirroring.  But what I'm wondering is this: if I use Raid 0 for striping, in this environment will their hard drive access be faster, slower, or about the same?  In other words, is it worth doing?  Or, should we just stick with a single IDE drive in the new server?

From all the stuff I've read, I get the impression that striping puts various pieces of their files across multiple hard drives, but I don't understand how the computer/operating system/controller--or whatever figures it out--could possibly know how to optimize what is placed where on which drive in order to enhance disk performance.  Does it somehow know?

Furthermore, I'm under the impression that the "system" will see a Raid 0 array as one drive letter, so I assume this means that we could Ghost an image of the "drive", i.e. the Raid 0 array, just like we do now with their single-hard-drive XP system.  But if we ever had to restore an image, is that even possible?  Would Ghost--or the computer/operating system/controller--somehow know how to restore the image properly across the multiple hard drives in the array?

And if the answer is "no", you can't restore a Ghost image, then what software is out there that would let you restore an entire image onto a Raid 0 array?

I'm so confused--obviously--and would appreciate any expert help!
LVL 3
sasllcAsked:
Who is Participating?
 
Duncan MeyersConnect With a Mentor Commented:
You will get speed benefits on read operations with RAID 5 - no question.

You will take a performance penalty on RAID 5 write operations - no question. The amount of performance penalty depends on the RAID card that you select. If you use a proper hardware RAID 5 controller with 256MB or more of cache, then you'll find that write performance for 10 users is very good.

The biggest benefits of RAID 5 are storage efficiency, redundancy, expandability and performance - you get plenty of disc space and you get excellent read performance.

The biggest benefit of RAID 1 is that you get good redundancy and reasonable performance (2 x single disc for reads, same as single disc for writes) - and its relatively cheap to implement.

RAID 5 is probably your best choice because: it is simple to expand (depends on the RAID controller - choose carefully!) - you add another disc, kick off an expansion and then increase the OS volume size (remember, though, that the C: drive in Windows cannot be expanded using the native tools - you need a 3rd party app.); it has good performance (again, depends on the RAID controller) and it has good redundancy.

For 10 users, go with RAID 5. If the company expands rapidly and needs more disc space, its a simple matter to provide it.

BTW - the cost thing is around needing a minimum of 3 discs for a RAID 5 set as opposed to just two for a mirrored pair (RAID 1). Also, a good RAID 5 controller will cost you more.
0
 
NightmanConnect With a Mentor CTOCommented:
RAID 0 would be faster BUT it is NOT fault tolerant. A single error on any of the drives will mean that you lose ALL data (except for backups - assuming these are stored elsewhere). Only use RAID 0 for non-critical tasks.

Good information here http://www.ibeast.com/content/tools/RaidCalc/RaidCalc.asp

Don't know about Ghost though.
0
 
NightmanCTOCommented:
And yes, the OS would be able to see the RAID array as one drive letter.
0
2018 Annual Membership Survey

Here at Experts Exchange, we strive to give members the best experience. Help us improve the site by taking this survey today! (Bonus: Be entered to win a great tech prize for participating!)

 
CallandorConnect With a Mentor Commented:
RAID-0 will help even on a single computer, so it will help on a networked drive.  The question is, are you suffering from an I/O bottleneck such that this will help?  If not, you are introducing an increased risk of loss of data for no apparent gain in access speed.  In other words, if you are not hitting the limits of a single drive access, moving to a RAID-0 array will not improve access.  I know that a typical IDE drive can deliver 50-60 MB/sec sustained transfer rate, and a WD 10,000 rpm Raptor can deliver 70 MB/sec.  In a RAID-0 configuration, two Raptors can deliver 100 MB/sec sustained, so it can benefit, if needed.

The RAID-0 array is "seen" as a single drive, so you can image it and restore from an image to an array, or restore even to a single drive - the software doesn't care.  Just keep in mind that everything on the array is lost if one drive fails.
0
 
netstableConnect With a Mentor Commented:
Maybe RAID0+1 would be the safest option striping with mirroring is a much safer way if not more expensive, a gigabit network might be a little more beneficial too :)
0
 
sasllcAuthor Commented:
Makes sense to me that with a RAID-0 "everything on the array is lost if one drive fails"--just like on a single drive, I assume.

And I'm under the impression that the best way to avoid that situation is to include mirroring, as well...which would be a RAID-10 configuration?  Because with RAID-10 I am striping across drives, but also mirroring those drives to keep me up in the event of a single drive failure.  All correct?

And finally, I get the impression that RAID-5 does striping, and also includes parity checking, and that this gives increased data integrity, like RAM parity does, but that RAID-5 alone cannot and does not give you any protection in the event of a drive failure, because it does not do any mirroring.  Correct?
0
 
CallandorCommented:
Mirroring a stripe (RAID-10) will help, but at the cost of half your disk space (you only get to use half of your total drives for anything).  RAID-5 is better for disk use, because you only lose the space of one drive, no matter how many you use in the array.  RAID-5 can survive a single drive failure; the array will be in a degraded state until you replace the failed drive, at which point you can rebuild the array with no data loss.  RAID-6 has arrived and it can survive two drive failures.  A good quality RAID-5 controller can produce performance on par with a stripe; for example, I have an Areca 1220 with 5 drives that can transfer data at 250MB/sec sustained on the outer tracks, going down to 150MB/sec on the inner tracks.
0
 
sasllcAuthor Commented:
So....it sounds like RAID-5 is the 'best of both worlds', because:

1) It uses more of the available disk space

2) Potentially provides for faster data access in some cases, even though it has to write parity data, because it stripes the data across drives

3) It still 'keeps going' even if one of the drives in the array dies.

Is that all correct?

How easy or hard is it to 'rebuild the array'?

Would it still make sense to make a Ghost image each night, that could be restored like you would restore to a single drive?

Can I count on a RAID-5 with a hardware controller being no SLOWER than a single, fast, non-RAID IDE setup in terms of overall disk performance (using the same server for either one)?

Finally, I'm still wondering if, in this small office environment I've described, if it would be just as good to get one super-fast IDE drive and Ghost it every night.  I do realize that I would not have real-time instant recovery and stay up and running in the event that my hard drive failed.  But if that is not a factor for this customer, then in that case I still wonder if it makes sense to go to all this trouble and expense for RAID of any type.  To my knowledge, they are not 'suffering an IO bottleneck', and I don't know of any 'scientific way' to find out.  (The only thing that has been consistently slow has been opening screens (forms) in Accpac, and to date, we have no idea where that is coming from or why it happens).  Instead, we're just trying to 'do the right thing' since they are making this change to a real server and adding Exchange server.

So, any answers to the above and/or additional recommendations based on the above are appreciated!
0
 
CiaranDolanConnect With a Mentor Commented:
1 - It requires a minimum of 3 drives and you lose 1/3rd of the total space - so, yes.
2 - On your network - you won't notice the difference.
3 - Yes

Rebuilding the array is normally a simple matter of replacing the failed drive. Either a hot-swap or a shutdown and a cold-swap.

Finally - no that is not a good idea. It's better to get a redundant system rather than  just rely on a ghosting. Add on a decent backup system then you will be set.

BTW, watch out for the complexity of Exchange. It's going to cause some issues!
0
 
Duncan MeyersCommented:
>Can I count on a RAID-5 with a hardware controller being no SLOWER than a single, fast, non-RAID IDE setup in terms of overall disk performance (using the same server for either one)?

Yes and no. A RAID 5 setup will generally out-perform a single disc on read operations, but it will be up to a quarter of the performance of a single disc for write operations (assuming no write cache).

As a rule of thumb, RAID 5 will provide (number of discs -1)*(single disc performance) for sequential reads, less for random reads.  Note that the rule of thumb goes out the window with multiple ATA/IDE/SATA discs, although it still applies for SCSI discs. You get what you pay for. SCSI discs are more expensive and they'll give better performance.

The reason you get poorer performance for RAID 5 write operations is the RAID 5 write penalty. For each RAID 5 write, the controller must perform 4 discs operations: read original parity, read original data, generate new parity, write new parity, write new data. The write cache on the RAID controller will help with the performance impact, but if you fill up the write cache, then you're down to the underlying disc performance.

For the environment you describe, RAID 5 will work well. There is absolutely no place for RAID 0 in a business environment. If you're constrained on cost, a mirrored pair of big SATA discs would be the way to go.

You still need to make nightly backups - there is no getting around that.
0
 
sasllcAuthor Commented:
OK...I'm really going to try and make this the last question on this subject...when you say 'If you're constrained on cost, a mirrored pair of big SATA discs would be the way to go', I assume you're talking about a RAID-1 setup.  But from what I'm reading I can't understand what the benefit of RAID-5 would be over RAID-1 for this environment...comments above such as 'On your network - you won't notice the difference' make me think that RAID-5 likely won't buy me any speed advantage over a single drive in this environment, and comments about the "RAID 5 write penalty" indicate that I may even lose some speed on writes, due to all the parity stuff--as compared to a single IDE drive.  

And, apparently I can get the mirroring/redundancy benefit by going with RAID-1, although I understand it has slower writes as well, because it's writing to two drives.

I get the impression that you're suggesting that RAID-5 is the better way to go if the cost is not an issue, but I can't understand why.  What is the benefit--in this environment--of RAID-5 over RAID-1?
0
 
Duncan MeyersCommented:
Glad to be of help!
0
All Courses

From novice to tech pro — start learning today.