RAID help needed

[michaelnotar]

just bought a lacie 2TB array consisting of 4-500GB HHDs w/ RAID. curious to know what level of raid i should use.

my choices that i'm considering are 0+1(stripping and mirror), 5 (parity), also available 5+ hot spare (dont care about the downtime getting a new drive in event of failure), and 0 (pointless for my purposes)

its currently set up in 5, but it is my backup device so mirroring seems more appropriate. i am also storing all my raw images also on DVD, and for important jobs/weddings/etc i will also burn another copy of all files on DVD. so thats: HHD array for everything; DVD for all raw and the ocassional important project. considering this what would you reccomend?

i've been reading up the raid level differences and i am curious to know if 2 drives fail of an array, if any raid can rebuild itself? probably not common...thus perhaps itseems the RAID levels worry more about the more common one drive failure. perhaps the only way would be if RAID 1 were used and drive #1 and #2 went down, leaving #3 (#1 copy) and #4 (#2 copy) which would be ok, but if both #1 and #3 went down, you're screwed, but i think this is theroetically rarer.

in some ways, the current way my drive is setup (Raid 5) seems to be quite good for security, sometimes i think its equal to raid 1 for this, in that data is broken down accross multiple drives, making multiple failures neccessary to take them all down.

also, sources refere to blocks of data on HHDs for RAID, curious to know how this info translates into english. will one block of data change an image in a way, perhaps a dead or odd colored pixel? will it corrupt it/make it unusable?

TIA for reading all this and your help.

[MikeA]

Without knowing exactly how your hardware behaves in failure mode, I can offer this insight:

If you have multiple drives (ie. more than 3) and if the sum of the drive capacities divided by 2 is more than you need, definitely go Raid 0 + 1.

For a backup drive, I wouldn't go Raid5. It will impact write performance. With Raid5, every write to the array involves controller calculation of parity and a write to every drive in the array. The suppliers of Raid5 will tell you that the controller has more than enough cpu capacity to handle the calculation (and maybe it has, maybe it hasn't) but the issue is multiple sequential writes to multiple drives, and I can tell you that cpu is not the issue as much as IO is. On top of that, if you talk to the data recovery people, you will find that there are significant corruptions of Raid5 arrays on single drive failures. Shouldn't happen, but it does. I could tell you stories that would make your hair curl...

So, if it were me, I'd go Raid 1+0 with a hot spare. How many drives can fail before we loose integrity of the drive? Well, that depends, but theoretically, with Raid 0+1, as much as half of them, but as little as 2 of them (3 if you have a hot spare, and the array has a chance to sync the hot spare before the second drive failure occurs)

If the data is real important, I'd also be looking at an offsite rotating storage system on portable media (DVD, Tape, Portable disks) The idea is to protect against user accidents and disasters of any kind, you cannot achieve that easily if the data is all in the one physical location.

Hope that helps,

Mike.

[Oyvind Nikolaisen]

It's easy, isn't it? :-)

At work, where I definitely don't do anything related to photo, rather more with networks, servers and SAN (Storage Area Networks), we have something like 3.000 (three thousand) drives total in our datacenters. All servers have been set up with Raid 1 for the OS disks, with Raid 5 for just about everything else. For heavily loaded database/Exchange server clusters, we set them up with either Raid 0+1 on DAS (Direct Attached Storage) or we connect them to one of our SANs, where the norm is a staggering 168pcs 15.000 rpm disks that share the I/O-load - the performance is unbelievable - and where the SAN controllers handle multiple drive failures transparantly to the attached servers.

For most uses where neither high availability nor performance is an issue, use Raid 5. Hot spare is nice if you have the space and $$ to spare for it, otherwise you'll have to live with a degraded array until you can replace the failed drive. One other consideration you may want to dvelve upon, is hot plug drives or not. Replacing drives may be a bit finger and screwdriver gymnastics unless they are installed in trays, I don't know how the Lacie products ship. Also, will you be able to replace a failed drive in four or five years time when the norm may be 1 TB drives at the lower end or solid state disks may rule the market?

With regards to performance on Raid 5 - it all depends on the controller. I'm used to high end server controllers where this really isn't an issue, motherboard controllers on cheaper PCs may have a number of disadvantages in terms of write performance.

Short version - use Raid 5.

Best regards
Øyvind Nikolaisen

[MikeA]

Øyvind,

We disagree. Love your country though.

Mike

[michaelnotar]

thanks for everyone's help thus far.

i really dont care much about data transfere speeds as i am not using them for a scratch HHD.

sounds like 0+1 is the way to go, why risk it...5 sounds preatty good, but i will error towards caution,and it seems this one is faster anyways.

i cant do 0+1+hotspare, the benefit of the hotspare is just down time? as it it sits there unused untill its needed to rebuild? so it saves me a day or two while a new drive is sent to me? is that all?

BTW, the drives are in bays, which are really easy to remove or replace.

[williamhenshall]

And here's my personal experience:

In my previous career up to 2003, I was Chairman and Founder of a high tech network company in San Francisco. We designed and built a client server system called "DigiDelivery", which is now an Avid/Digidesign hardware server product that provides super fast super reliable large media transfers over the public internet. It's like FTP but it works every time with larger files (ftp is unreliable with files over 200 megs) Here's a link to it in case you're interested.

The system is used today in many Hollywood movie studios for audio post production, XM radio, NPR radio, the big 4 broadcast companies in the usa and globally use this to push work in progress data files around - up to 20 or 30 gigs sometimes...

We also had a data center that provided this service to clients as a pay per file offering in a bunker in San Jose - the initial total server equipment investment for this was about 15 million usd.

I mention this because this gave me much hard learned experience with large scale backup systems and how to design and manage raid arrays. And we ended up exclusively using raid level 5 arrays running on SAN networks. But we used super expensive dedicated hardware systems from Sun and Adaptec and also had a few of the then new Apple X-Raid boxes.

However there is a big difference between enterprise mission critical banking level NAS systems running level 50, 500 and 5000 (ie nested sets of level 5 drives that appear as a single drive to the user) and what photographers need on a local level...

When any one asks me what raid system to get my instant answer is get the Apple X-Raid. This system is unlikely to ever let you down and integrates using fiber into your network. It is screamingly fast and remarkably easy to set up, and can support many users. It truly rocks.

But it's $13k for a 7 Tbyte system. However, if ya want a Rolls Royce that ain't gonna breakdown, there it is. I'd love to justify having one here on my own network. Heh.

Anyway, reality beckons and here in my own little studio in LA, I still use level 5 NAS raids, by Snapserver and also by Caen, and we keep an off site back up of all files. Although with the event of Lightroom I have started just archiving the LR database and the RAW files from each job together with any specific PS CS3 versions and nothing else.

I have not personally ever had a non recoverable disk crash from a level 5 array. Which of course is not to say it can't happen...just that it has not happened to me

Here's a link to an excellent explanation of the different RAID options

In my view the best way to work is:

Each photographer runs an internal mirrored RAID 1 for work in progress data in his/her workstation machine (this gives 2 automatic copies) and then each days work is archived on to a networked level 5 server at the end of the day and then every week we do an off site safety copy of the new work. This makes a single burden of an end of the day copy the only task for our photographers. Once the job is on the server it also means that anyone else can access the job as needed and our IT guy does the rest of the archiving from there.

Depending on how much the client has paid us for the job (heh, true...) we may do mirrored or non mirrored off site back up. That's more to do with my comfort level than anything else! If the shoot involved helicopters, yachts, bunny girls and a couple of fortune 100 CEOs, then hell ya we gonna get those btyes mirrored off site!

One thing that is worth mentioning is this - reliability of drives is affected most of all by sudden power outages. So the most important thing you should do is run quality UPS power systems with auto shut down on all workstations and servers. It's not a question of if you get a power surge or cut, it's a question of when. And this is when you will get disk corruption on any kind of drive set up!!!

Enough high tech discussion for now, back to the previously mentioned bunny girls...

Will

[MikeA]

Michael,

The benefit in the hotspare is in minimising the risk to the array of a second drive failure. So, it's a risk mitigation and time saver. In the event of a drive failure, the raid controller brings the hot spare online and synchronises the data with no input from the user.