• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

RAID 5 in production server

S

snappahd

Junior Member
I'm looking to build an inexpensive 1.0TB NAS box using inexpensive IDE hard drives. I am intending on building a xeon class PC with a rack mount chassis, lots of fans and 5 250gb Western Digital SE hard drives using Serial ATA and a 3Ware 8 port SATA controller.

Has anyone had any experience in this realm? I am wondering how well these solutions recover from a drive failure? Any gotchas I should keep in mind, or are there any tips that you have learned from your own experiences?

Thanks!
 
Originally posted by: shuttleteam
Experience dictates the use of SCSI disks in any production environment where uptime is a concern.

-DAK-
Click to expand...

Why? Is the MTBF for SCSI higher/longer than for IDE? If cost takes precedence over speed (which SCSI has the definite advantage) I would think IDE drives would be ideal. It's cheaper to get a few large IDE drives than a lot of smaller SCSI drives.
 
Why? Is the MTBF for SCSI higher/longer than for IDE? If cost takes precedence over speed (which SCSI has the definite advantage) I would think IDE drives would be ideal. It's cheaper to get a few large IDE drives than a lot of smaller SCSI drives.
Click to expand...

MTBF is just a published number. IDE drives are built like crap these days and are failing a lot more than SCSI disks. Of course it's far cheaper to use IDE. What's the cost of downtime? Also, no IDE HBA has the online features of Enterprise Class SCSI HBA's. NONE!

-DAK-
 
I"m currently running a 3ware 4710- (4-port raid 5 with 4x80gb) and i'm extremly happy with it. I messed around with it some before I actually put valuable data on it, and it's not that hard to recover. I built a raid 5 array, then filled it partly with misc. stuff. I then took one of the Hard drives out, and put a clean one in it's place. I booted into the raid cards bios, and simply rebuilt the array using the new drive. It worked like a champ, and i must say I'm quite pleased with 3ware. I just purchased a used 3ware 6400 (8-port) and I plan on filling it with 80 gigs too.

-Chris
 
Depends on how important your data is.

If it's something like a database, etc, I definately agree with shuttleteam, go SCSI.
If it's stuff that's not vital, and you need alot of cheap space, S-ATA RAID5 would do fine IMO.

At work, we're considdering a similar setup for temporary storage of alot of alot of junk that doesn't have to be available all the time, if it goes down for a while, it won't be no biggie.

Depends entirely on the situation.
 

We were looking into this as well. Something around a 300-500gig solution. For my storage needs I priced it up and I could build one from newegg for about 1500 bucks. We also looked at Dell's. Their NAS solutions, although competitive, are much more expensive. We settled on a low end Dell Server stacked full of ATA drives. For a ~350 gig solution it runs about $2500 including enough drives to keep the OS on a separate Raid 1 array. Building it yourself is cheap but 2hr-response, 24/7 support is worth the money if it's in a production environment.

It depends on the environment of course, but SCSI is not a requirement for production servers any more. To answer someones question above though: Yes, SCSI has a MUCH higher MTBF than ATA. The catch is that the MTBF of ATA is long enough you'll probably replace the server for other reasons before you have a failure. As always my law on this stuff is, "Your Job is only as secure as your last backup."

As for processor: if it's just going to be file serving you don't need a Xeon processor. I've got a 1/2 TB NAS here serving about 1200 users with a single 1gz PIII and it never breaks a sweat - I don't even think I've seen it hit 20% during morning logons.

The other thing to think about when looking at NAS is licensing. With our solution above we settled with Windows 2000 for the OS so we needed microsoft CALs for every connecting user. If you get a "real" NAS it will come with an appliance OS - typically a "built on Windows 2000 technology" operating system. These OS's don't need CALs. If you're already someplace that has enough CALs for everyone this isnt' a big deal. If you need to buy cals to connect the cost is going to make a "real" NAS look very attractive.

 
SCSI was built for 24/7 uptime and made for enterprise environments where data is key. IDE drives were meant for desktop use, where someone pulls the plug from them on a regular basis. If you want something reliable, in a RAID-5 situation, I would recommend getting SCSI.

Here's another thought on the IDE/SCSI for RAID. For many SCSI drives, you can get *nearly* the same exact drive 2-3 years down the line, with roughly the same specifications. For IDE, that is much more difficult. For RAID situations, its always better to get matching drives. SCSI makes this easy, IDE can make this difficult.

vash
 
Thanks for the great info. The intended application is for a backup solution. Since tapes are so expensive still, and hard drives so cheap, I wanted to create a 'big disk' and use veritas backup exec to write my backup sets to it. v9.0 now supports NAS's as a backup destination, so I thought this would work well. This will largely be used for 'convenience' backups and restores, and my most critical backups that are off-site will still be using tape.

 
Originally posted by: vash
SCSI was built for 24/7 uptime and made for enterprise environments where data is key. IDE drives were meant for desktop use, where someone pulls the plug from them on a regular basis. If you want something reliable, in a RAID-5 situation, I would recommend getting SCSI.
Click to expand...

Yes, I think this is the original thinking that most of us grew up with, however, I think more creative thinking can net your company solid solutions using different technology at substantial cost savings. I believe as long as the IDE raid card you get can rebuild a RAID set reliably and quickly after a failure, you have all you need.

Here's another thought on the IDE/SCSI for RAID. For many SCSI drives, you can get *nearly* the same exact drive 2-3 years down the line, with roughly the same specifications. For IDE, that is much more difficult. For RAID situations, its always better to get matching drives. SCSI makes this easy, IDE can make this difficult.
Click to expand...

With the cost savings of my IDE solution, I am intending on purchasing an extra drive or two when I make my original purchase. The drives will be matched and I'll have plenty of spares even when I EOL this device in a few years. Maybe my then SCSI will catch up to IDE's Price/Performance.
 
SCSI will never catch up with IDE in terms of price/performance.

They are for different things, SCSI is the way to go when reliability and performance are paramount, but if you can afford to compromise, IDE has indeed become a viable alternative.
 
Since RAID 5 is fault-tolerant, does it matter if IDE drives are really less reliably than SCSI drives? Seems like if one of the IDE drives fails in RAID 5, you just replace it with another. Remember that the I in RAID stands for Inexpensive. 😀
 
Originally posted by: snappahd
Thanks for the great info. The intended application is for a backup solution. Since tapes are so expensive still, and hard drives so cheap, I wanted to create a 'big disk' and use veritas backup exec to write my backup sets to it. v9.0 now supports NAS's as a backup destination, so I thought this would work well. This will largely be used for 'convenience' backups and restores, and my most critical backups that are off-site will still be using tape.
Click to expand...

I recently did exactly this for a small company I do contract work with.

I bought four WD 200GB SE's. Used a 3Ware 7500-4LP card and created a 600GB backup array where we can dump the daily backups. Then once a month run a tape backup of the most recent data. Working like a champ. The 3Ware cards are the only cards I have seen that have features close to SCSI Raid cards. It has a web interface, monitoring, and on-the-fly live rebuilds. Here's the best part: Hot Swap IDE. Had one drive fail on me (didn't really die, was a problem with the firmware and can be seen here), I Hot Swapped a new one in and rebuilt it; all without restarting the machine. This is as close to SCSI reliability as IDE can get, and it cost thousands less.
 
Originally posted by: vash
SCSI was built for 24/7 uptime and made for enterprise environments where data is key. IDE drives were meant for desktop use, where someone pulls the plug from them on a regular basis. If you want something reliable, in a RAID-5 situation, I would recommend getting SCSI.

Here's another thought on the IDE/SCSI for RAID. For many SCSI drives, you can get *nearly* the same exact drive 2-3 years down the line, with roughly the same specifications. For IDE, that is much more difficult. For RAID situations, its always better to get matching drives. SCSI makes this easy, IDE can make this difficult.

vash
Click to expand...

1 Terabyte of Space = 14 x 73Gig SCSI drives. You won't be able to fit these on a single channel if you throw an extra drive for parity into the mix so you'll end up with (7 + 1) x 2 = 16 Drives.

A quick check at newegg... Fujitsu 73GB, 10krpm = $349

1 Terabyte of Space w/ parity = 5 x 250Gig ATA drives.

Another quick check at newegg... Maxtor 250GB, 7200rpm = $262

Total Cost for SCSI = $5584
Total Cost for ATA = $1310

I don't have the MTBF numbers in front of me but you could replace every single drive in the ATA setup four times over and still cost less than SCSI. "But what is the cost of your downtime!", you say? Dunno. He'll have to make that call. SCSI might be more reliable but what's the chances of having a drive fail when you have five of them? What are the chances a drive will fail when you have 16 of them?

SCSI has it's place for sure, but this ain't it.
 
1 Terabyte of Space = 14 x 73Gig SCSI drives. You won't be able to fit these on a single channel if you throw an extra drive for parity into the mix so you'll end up with (7 + 1) x 2 = 16 Drives.

A quick check at newegg... Fujitsu 73GB, 10krpm = $349

1 Terabyte of Space w/ parity = 5 x 250Gig ATA drives.

Another quick check at newegg... Maxtor 250GB, 7200rpm = $262

Total Cost for SCSI = $5584
Total Cost for ATA = $1310

I don't have the MTBF numbers in front of me but you could replace every single drive in the ATA setup four times over and still cost less than SCSI. "But what is the cost of your downtime!", you say? Dunno. He'll have to make that call. SCSI might be more reliable but what's the chances of having a drive fail when you have five of them? What are the chances a drive will fail when you have 16 of them?

SCSI has it's place for sure, but this ain't it.
Click to expand...

$5k is nothing. Time is valuable. The risk of having downtime at the cost of $100k per hour is unacceptable! If this applies to you, the answer is obvious.

Does the 3Ware support volume expansion of logical volumes? I've actually run out of space (well come close to it) on a running system and it was just a matter of adding two more disks and expanding the capacity of the R5LD. Everything was done in the background with z-row downtime! If ATA can do THAT, I'm all ears...

-DAK-
 
To add to Smilin?s post about CALs, if you attach this to a 2nd NT/2K server and you change your licensing mode to per seat instead of per server, you wouldn?t need any more CALs in your environment. You would however have to pay for another server license. If you are going to do this as backup instead of NAS though, wouldn?t XP Pro work fine? I know it has a limit of 10 network connections or something like that, but you?d only have one or two from your backup software. Wouldn?t XP Pro be robust enough for that? What?s everyone?s thought on that?

I have a customer that is going to backing up like this to another building a mile away (connected via fiber). I figure it would take quite a fire to knock out both buildings and destroy the backups. It looks like we are deciding to go with a Promise UltraTrak RM15000 with probably 8 WD drives to start with. I?m not sure how easy it will be to add more drives later. Still researching that. Promise notes that ?The SCSI-2 specification limits the logical drive (array) maximum size to 2 Terabytes. For larger capacities, multiple arrays can be used.? That surprised me. We need about 3TB. Could get real interesting. Maybe we will start with a single RM8000 and add a 2nd one later when we need to expand. You can check em out here:

http://www.promise.com/product/product_list_eng.asp?familyId=6
 

*bing* lightbulb went off.

Yeah, XP pro would be a perfect OS for something like that. I think they've bumped it from 10 connections down to 5 starting with XP but if you're just backing up to it you're only gonna use one.
 
Down to 5 now huh?! More robust OS, less connections, I'm not surprised really. Can't have people NOT paying for those server costs... Would XP Pro stream as fast I wonder as Server? I know Server is more built for services such as file sharing and printing, etc. I wonder if there would be any appreciable speed difference between Pro and Server over 2Gbps LAN and to Ultra320 hard drives. (I'm thinking about a 2nd customer now that was going to put in a 6th Win2K or 2003 Server primarily for backing up to.) Just wonder if one might be faster enough to shave time off the noon backup and make it fit better into the lunch break. I may have to do a couple OS loads on it and measure it just for kicks before purchasing a legit license... ;-)
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top