RAID setup recommendations

[Brian Skalinder]

I'm planning on putting together a RAID-1 or -5 setup to store all my audio (master recordings and commercial releases) in FLAC, ultimately for playback around the house with something like the SlimDevices Squeezebox.  So...any recommendations on:

• PCI RAID controller
  
• external HDD enclosure for 2 or 3-5 HDDs (depending on whether I go with RAID1 or 5), incl. power supply, does not need to support hot-swap
  

I think for my purposes, i.e. limited I/O activity v. enterprise-level constant I/O hammering, ATA/SATA are just fine.  If a drive bites it, I'll just take everything down, replace it, and re-build the array.  So no need to go with the more robust SCSI unless someone convinces me otherwise.  It's been so long, not sure what else I need!  Cabling, of course, but I figure that's a no-brainer.

I'm looking to keep this fairly low-cost, but in the arena of Good Value (low cost / strong performance) rather than Cheap (low cost / crap performance).

Recommendations?  TIA.

[OFOTD]

Stick only with Promise cards.  You can buy others but you'll end up going to Promise cards.

I use the FastTrak TX2000 card.  Its SATA/RAID.  They do have a similar card that is IDE but if you are starting fresh I would recommend SATA.

I think I bought mine for $95 I think.

http://www.promise.com/product/product_detail_eng.asp?segment=undefined&product_id=88

it only does Raid 0, 1 & 0+1 though.

[it-goes-to-eleven]

Good move.. I need to go RAID but haven't yet.  I've let it go Way too long.  Now, a hunger for dime has left me with a big problem that is growing faster than drive capacities.  My cases are full of drives and I need to go external..  It will be tough for me to migrate to raid. Save yourself before it is too late!

I've had bad luck with IBM and Seagate and now only use Western Digital (for the past 3-4 years). They also have Advanced warranty replacement (they ship the drive to you first) and it has worked well. Seagate did not offer that the last time I did an RMA with them (20 months ago).

Make sure you get OEM drives which include 3+ year warranties. Retail are now only 1, with an upgraded warranty costing around $15 (too much).  Plan on monitoring your drives for health status (check drive temps and run offline diags every 6 months or so). Most of my drives report running below 30C and sit behind a fairly low speed 120mm case fan.  Drive temps in regular cases are usually well over 40C.

I worry more about theft and fire/etc.. So that means some kind of off-site solution (at least for stuff that cannot be replaced).

I think raid 1 is overkill unless it includes good support for off-site storage...  Maybe raid 5 at home (or no raid and automatic nightly snapshot backups?) and an external drive enclosure you store elsewhere and use to snapshot the home system once a month or quarter?

FWIW, I just yesterday ordered a 320 SATA western dig from monarch for $135 shipped.. And the order status shows that they upgraded my free ground fedex to 2 day. They've been beating newegg by quite a bit lately. SATA prices seem to have fallen below IDE for the larger drive sizes.

This is the best computer case I've found for having a lot of drives and keeping them cool. It will hold 5 3.5" drives comfortably behind a 120mm fan (not included) in a nice removable cage (plus 2 more if you use the floppy bays). I have 5 drives in two of these cases and most of them run below 30C:

http://www.newegg.com/Product/Product.asp?Item=11-129-152&depa=1

[BC]

Plan on monitoring your drives for health status (check drive temps and run offline diags every 6 months or so).

Dumb question: how do you do this?

[Busman Audio]

go with the Raid 1 because the expense in a RAID 5 will be higher. Also, if you are not using the system as  a DAW to do multi track recording you don't need the speed of RAID-5. SATA will do fine again SCSI is paying more for speed you may not use.

[Brian Skalinder]

go with the Raid 1 because the expense in a RAID 5 will be higher.

It's only one more drive expensive though, and it provides more storage capacity for the buck.  Though I suppose the controller might be minimally more expensive as well.

Also, if you are not using the system as a DAW to do multi track recording you don't need the speed of RAID-5. SATA will do fine again SCSI is paying more for speed you may not use.

I was thinking RAID5 in part because it provides more storage capacity for the buck.  And thanks for the confirmation - no real need for SCSI speed.

[fozzy]

go with the Raid 1 because the expense in a RAID 5 will be higher. Also, if you are not using the system as  a DAW to do multi track recording you don't need the speed of RAID-5. SATA will do fine again SCSI is paying more for speed you may not use.

Depending on the situation RAID5 can decrease performance.  I don't want to go into the math but for our purposes of Large Sustained writes(saving large wavs) and Large Sustained reads(dither, wav processing, etc..)/short random reads(playing files, tracking shows)  RAID1 is what you want if the data is important to you.

Brian: blow your wad on the RAID card and get what ever drives you can afford (i recommend something in the 150-250gb range which seems to be the current price point).  Can't go wrong w/ the promise cards.  SATA/ATA/SCSI: I'm happy w/ all the RAIDs I have except for the heat the SCSI generates.  Any performance differences will be minor between the technologies, especially w/ a dedicated controller.

[nic]

Brian, is this RAID setup strictly for listening/storage or for archiving?
R1 is mirroring which gives fault tolerance.
R5 is stripping with parity which really doesnt give any fault tolerance and does not give any additional storage space over R0(regular striping). sounds like you might want to look at R0+1...granted, this requires 4 drives, but gives you the additional storage of R0 plus the fault tolerance of R1.

I never had that good of luck with the Promise cards, but I only used their standard ATA versions. maybe their SATA ver is better?
I currently have 3 R1 setups based on Silicon Image chipsets and they have all worked flawlessly...

[it-goes-to-eleven]

R1 is mirroring which gives fault tolerance.
R5 is stripping with parity which really doesnt give any fault tolerance and does not give any additional storage space over R0(regular striping).

But raid 5 does provide fault tolerance.

The parity data that is calculated and retained as part of raid 5 is the redundancy. It is used to reconstruct the contents of a single failed drive (whether on the fly or when writing to a replacement).

[nic]

yes, the parity drive does provide 'some' fault tolerance, but it is not nearly as easy to rebuild a failed array as it would be with RAID1 or RAID0+1

[fozzy]

yes, the parity drive does provide 'some' fault tolerance, but it is not nearly as easy to rebuild a failed array as it would be with RAID1 or RAID0+1

I would disagree.  Most scenarios of failure are a single drive failure.  In both schemas multiple drive failures result in complete data loss.  Rebuilding from parity or rebuilding from a mirror is generally seamless and can happen while the system/RAID is active.  They can also operate in a non-optimal(failed disk) state w/o a performance impact.  The only advantage i can see is that in a pure RAID1 setup (not RAID 0+1) it is sometimes possible to break the mirror and mount a single drive in the case of a failed or corrupt RAID configuration. 

[it-goes-to-eleven]

There's a huge difference between what raid 5 provides and 'doesn't give any fault tolerance'.

And, how often are you dealing with a failed drive in this application?  Since this isn't a data center with thousands of drives.

Billions and billions of dollars in data are protected with raid 5.

[Brian Skalinder]

Brian, is this RAID setup strictly for listening/storage or for archiving?

Both.  I'll use HDD as my primary storage for listening and as an archive.  I'll continue to archive my masters to DVD.  And, I'll continue to ensure I have off-site backup by distributing the recordings to the broader music-loving community.

yes, the parity drive does provide 'some' fault tolerance, but it is not nearly as easy to rebuild a failed array as it would be with RAID1 or RAID0+1

I would disagree. Most scenarios of failure are a single drive failure. In both schemas multiple drive failures result in complete data loss. Rebuilding from parity or rebuilding from a mirror is generally seamless and can happen while the system/RAID is active. They can also operate in a non-optimal(failed disk) state w/o a performance impact. The only advantage i can see is that in a pure RAID1 setup (not RAID 0+1) it is sometimes possible to break the mirror and mount a single drive in the case of a failed or corrupt RAID configuration.

I'm inclined to gree with you, Fozzy - my old IT background is slowly coming back to me, and you've mirrored my understanding.  About the only reason I can think of for why I'd want to run RAID1 over RAID5 is your final comment:  the ability to mount a standalone single drive in the event of drive failure or corruption.

I initially wanted to run RAID5 in part so I could increase storage capacity easily:  just add a drive and - BANG - X GB more storage capacity.  I could increase capacity in RAID1, but it wouldn't prove as simple.  I'd have to either:

[1] create a second array
This would require another controller (or two more if I'm duplexing - don't think that's necessary, though) and 2 more drives as opposed to the ability to add single drives as necessary with RAID5

[2] increase array capacity
This would require a bit more hassle...

• add two larger drives to the array
  
• re-build the mirror to include the new, larger drives
  
  • which, of course, won't utilize the full capacity of the new drives (yet)
• remove the old, smaller drives from the array
  
• increase the size of the mirror to utilize the full capacity of the new drives

Option [2] doesn't make such sense to me - unless for some reason I don't want or can't accommodate multiple RAID1 implementations - since I'd have the smaller drives leftover doing...nothing.  Might as well go with [1] and build a whole new array and maintain the storage capacity from the original array + the new capacity from the new array.

Sounds a helluva lot easier to just chuck a new drive into a RAID5 array!

[fozzy]

I believe this is a false assumption unless you got some crafty software:
I initially wanted to run RAID5 in part so I could increase storage capacity easily:  just add a drive and - BANG - X GB more storage capacity.

RAID1 vs. RAID5 from an io cost-benefit

(3) Both RAID5 (parity) and RAID1 (mirrors) offer fault-tolerant disk storage. Since IOs are the scarce resource, one wants to optimize for IOs rather than for space. The  argument in favor of mirrors versus RAID5 is that mirrored disks offer double the read bandwidth to each data item, and they cost only one extra access for a write. RAID5 uses up to four disk accesses to do a write, and improves read bandwidth only if the data requests go to different disks. RAID5 saves disk space (gigabytes) at the expense of more IOs for disk writes.

[Brian Skalinder]

I believe this is a false assumption unless you got some crafty software:
I initially wanted to run RAID5 in part so I could increase storage capacity easily: just add a drive and - BANG - X GB more storage capacity.

Ahhh...I'm mistaken, then.  I thought it was possible (and reasonably simple) to add additional drives and re-build the array using the new drives.  Hmmmmm...gonna have to re-think this one, then.

RAID1 vs. RAID5 from an io cost-benefit

(3) Both RAID5 (parity) and RAID1 (mirrors) offer fault-tolerant disk storage. Since IOs are the scarce resource, one wants to optimize for IOs rather than for space. The argument in favor of mirrors versus RAID5 is that mirrored disks offer double the read bandwidth to each data item, and they cost only one extra access for a write. RAID5 uses up to four disk accesses to do a write, and improves read bandwidth only if the data requests go to different disks. RAID5 saves disk space (gigabytes) at the expense of more IOs for disk writes.

I'm not terribly concerned about IO performance since I'm not going to be hammering away at these suckers constantly with massive writes - only occasionally while editing audio.