To JBOD or not to JBOD, that is the question.

[skeet]

Thanks Tea & Cliptin (regarding 15,000 RPM)

The Storage Forum comes up trumps again.....

Just as an aside, has anybody had any experience with a JBOD (fibre channel). We are running three (all about 800 Gig). Is redundancy the only major difference between it and a RAID? We currently use them as video servers and they have been very effective in this capacity where the information they contain is important but replicable (re-digitise/ize (ha ha!).

We are exploring new storage possibilities and RAID has been used extensively in the past by proprietary systems when large storage arrays were in their infancy (notably Quantels Harry/ Henry/ Editbox/ Inferno, systems costing in excess of £300,000 for entry level). Is JBOD a price/ performance alternative when redundancy becomes less critical?

This may have warranted a new posting, in fact I will do it now...

Ps. The locals over here could do with some leather and willow assistance! (I'd help, but too I'm to busy LMAO collecting £££'s) :rofl:

 

[Cliptin]

Thanks Tea & Cliptin (regarding 15,000 RPM)

The Storage Forum comes up trumps again.....

Just as an aside, has anybody had any experience with a JBOD (fibre channel). We are running three (all about 800 Gig). Is redundancy the only major difference between it and a RAID? We currently use them as video servers and they have been very effective in this capacity where the information they contain is important but replicable (re-digitise/ize (ha ha!).

We are exploring new storage possibilities and RAID has been used extensively in the past by proprietary systems when large storage arrays were in their infancy (notably Quantels Harry/ Henry/ Editbox/ Inferno, systems costing in excess of £300,000 for entry level). Is JBOD a price/ performance alternative when redundancy becomes less critical?

This may have warranted a new posting, in fact I will do it now...

Ps. The locals over here could do with some leather and willow assistance! (I'd help, but too I'm to busy LMAO collecting £££'s) :rofl:

The advantage of JBOD vs. just stuffing a bunch of drives in a case is the ability to talk to the JBOD as a single unit. You could use mount points to approximate JBOD, too. I may be wrong but I don't think you can just tack another drive onto the JBOD when you run out of space either. I'm not sure what happens when you lose one drive of seven in a JBOD. Of course in RAID0, all the data is lost.

The closest RAID level approximation to JBOD would be RAID0. There is no difference between the cost of JBOD and RAID0 other that the SCSI card in the case of hardware RAID. This provides NO redundancy (repeat, not redundant). Among the RAID levels 0 is the fastest in streaming data.

 

[Tea]

I really really hate it when I type three-quarters of a long reply and when I look up at the screen again my browser window ain't there anymore. I could retype it from memory, but it's never the same second time around. WTF happened?

 

[Fushigi]

It's my understanding that using JBOD involves filling up each drive before writing to the next. Comparing that to RAID 0 (striping), RAID 0 should perform much better as disk accesses are spread across the drives more uniformly.

Expanding RAID 0 arrays will depend on the controller; some may or may not do it gracefully. It's my understanding that a JBOD array can be expanded by just adding the new unit(s) to the array.

In either scenario, all data is more than likely lost in the event of failure of any unit in the array.

If you are reading more than writing, don't count out RAID 5. RAID 5 also has very fast read performance however write performance is slower because of parity protection overhead. Still, with RAID 5 you do gain the ability to survive disk unit failure. The 'penalty' is devoting 10% - 33% of your capacity to RAID 5 overhead, depending on the number of drives in the array. Basically, for n drives where 3 <= n <= 10, you will have n-1 drives worth of capacity.

It is my firm belief that any business server needs data protection. It's a fundamental aspect of system reliability & availability. If you do not opt for a level of RAID that can withstand the loss of at least a single disk unit, then I hope you have a recovery method that is complete enough and fast enough to not severely impact the services that server provides.

FWIW, between my two main servers I have 95 18GB 15K RPM drives (IBM & Seagate). I'll be adding another 30 drives after the first of the year. We suffer a disk failure approximately once every 6 months (rough average). But because we use RAID-5, the system does not go down & the users do not experience any service interruptions.

- Fushigi

 

[Tea]

Fushigi, if I translate that rough figure of yours into the way that I think about reliability, which is drives per 100 fail inside the three year warranty period (for me, three years is the period, I know that various drives differ), I get 6 failures per 100. This surprises me!

I expect to get that sort of failure rate, perhaps a little higher than that, out of a typical "good average drive" - let's say a Western Digital 20BB or a Barracuda ATA. "Really good average drives" - some of the old Seagate Medalists or Dekstar 3s, for example, do around that well (6%) or better. And outstanding drives such as the Spinpoints or the 1.08GB IBM Deskstars do substantially better than that.

Now I would have imagined that the failure rate of, say, a Cheetah X15 or a 36LP would be in the order of two failures per hundred per three years. I should imagine that I have only sold or owned perhaps 30 top-line SCSI drives but if we discount a single Cheetah Mark 1 DOA and the inevitable Micropolis (which was hardly top-class anyway), I have never seen one fail. How do you account for your failure rate? Have I been very lucky? Or are the failing drives mostly the IBM ones? Or do you ascribe it to the much heavier workload that your drives have?

(PS: the drives I own myself all run 24x7, but those I sell mostly do a more typical deskyop workload, run maybe eight hours a day, and mine are spinning but idle for most of the time.)

 

[Tannin]

The locals over here could do with some leather and willow assistance!

Tell you what: I volunteer me and my boys to play for Australia. Um ... let me see, I'm 43 and the youngest player on the team. Most of the guys are in their late fifties, a couple of them well over 60. We ... um ... well, we finish second quite a lot in our third-grade Masters competition. (Second in the matches, I mean, not second on the table.)

But unless the ACB choses to take me and the boys up on this offer, I don't actually know what England are going to do about winning the Ashes back. Could they send out a Trident submarine or something?

 

[Handruin]

Thanks Tea & Cliptin (regarding 15,000 RPM)

The Storage Forum comes up trumps again.....

Just as an aside, has anybody had any experience with a JBOD (fibre channel). We are running three (all about 800 Gig). Is redundancy the only major difference between it and a RAID? We currently use them as video servers and they have been very effective in this capacity where the information they contain is important but replicable (re-digitise/ize (ha ha!).

We are exploring new storage possibilities and RAID has been used extensively in the past by proprietary systems when large storage arrays were in their infancy (notably Quantels Harry/ Henry/ Editbox/ Inferno, systems costing in excess of £300,000 for entry level). Is JBOD a price/ performance alternative when redundancy becomes less critical?

This may have warranted a new posting, in fact I will do it now...

Ps. The locals over here could do with some leather and willow assistance! (I'd help, but too I'm to busy LMAO collecting £££'s) :rofl:

See now I'm confused. At work we have a storage unit called Clariion (fibre channel disk array). This unit supports JBOD along with various raid levels (even both at the same time). My understanding of JBOD (just a bunch of disks) is that, if your array consists of 10 x 80GB drives, you can allocate 1 drive to a host which is 80GB unlimitly using just one physical drive. Therefore you do not need to allocate 1 drive which is 800GB.

With this, you could then assign 2 80 GB drives to a host and then do as you please with them independantly.

Now if you are using, say 5 drives for a total of 400 GB, and one fails, I believe you may suffur dataloss, much the same as running a spanned volume in windows. (However I don't know 100% that data will be lost if the drive which fails has no data on it)

I believe the difference between them is RAID. If JBOD is support in your storage array, that means you can access each individual disk independantly. Where as a RAID device will display as one physical drive to the host.

I think JBOD is definitely a price/performance alternative, but it depends on the type of storage array. For example, you get either with a Clariion, so it is up to the user how the space is divided.

Ive read for video that RAID 3 (Striping with parity) can help. The one example I've read says that the drives spindles are in sync and it is good for write-once, read-many.

 

[Cliptin]

I'm 43 ...

Happy Birthday.

 

[Fushigi]

Fushigi, if I translate that rough figure of yours into the way that I think about reliability, which is drives per 100 fail inside the three year warranty period (for me, three years is the period, I know that various drives differ), I get 6 failures per 100. This surprises me!

I expect to get that sort of failure rate, perhaps a little higher than that, out of a typical "good average drive" - let's say a Western Digital 20BB or a Barracuda ATA. "Really good average drives" - some of the old Seagate Medalists or Dekstar 3s, for example, do around that well (6%) or better. And outstanding drives such as the Spinpoints or the 1.08GB IBM Deskstars do substantially better than that.

Now I would have imagined that the failure rate of, say, a Cheetah X15 or a 36LP would be in the order of two failures per hundred per three years. I should imagine that I have only sold or owned perhaps 30 top-line SCSI drives but if we discount a single Cheetah Mark 1 DOA and the inevitable Micropolis (which was hardly top-class anyway), I have never seen one fail. How do you account for your failure rate? Have I been very lucky? Or are the failing drives mostly the IBM ones? Or do you ascribe it to the much heavier workload that your drives have?

(PS: the drives I own myself all run 24x7, but those I sell mostly do a more typical deskyop workload, run maybe eight hours a day, and mine are spinning but idle for most of the time.)

My X-15 runs 24x7 as well and keeps busy writing Folding results (in addition to the little green men).

For our data center environment, heat is not a factor so I can't blame that. I would say there are a few things happening:

1. Vibration. There are 30+ drives in a chassis + fans in a room with either a bunch of other similar systems (our prod environment) or 100+ Wintel servers (dev environment) plus HVAC & UPS units. Vibration from the chassis equipment + ambient vibes from everything in the vicinity may take their toll over time.

2. What exactly constitutes an error. A single miss in the parity or a soft error is enough for me to call a drive bad & replace it. With a production database accessed by 1300+ users, I don't take chances with data integrity (or my career). The vast majority of the time these are not hard failures/head crashes. I don't think we've had an actual hard failure of any of the 15K drives so far. Three or four on the 10Ks but none on the 15Ks so far. The soft errors can happen for a variety of reasons.

3. Workload. There are times when the drives are being pounded on. Take a 9GB data file and build 4 indexes against it concurrently. Watch disk arm utilization go through the roof. Especially since we're RAID 5 & have the extra overhead on disk writes. I could ease that & speed the index builds, but they probably don't want to drop $30K on extra RAM to buffer everything.

- Fushigi

 

[skeet]

Yikes Fushigi, I need more space. I've got GAS (Gear Acquistion Syndrome).
Thanks for your responses everybody, I'm about to attempt my first gear change as I ascend the learning curve.

Video of course differs slightly to true computer data because it is nearly always acquired by another device (camera) and stored on video-tape which is a remarkable data storage device. In fact one of our video formats is slight variation on a tape-based storage format devised for NASA (Ampex - DST/DCT, it's what killed them I'm told). OK, it's a bit slow with access times sometimes running into a minute, but where else can you a terrabyte for under a tenner.

As for the Ashes, the Trident could certainly deliver you some of those.

 

[NRG = mc²]

Hmm, looking at skeet's site, theres someone named Tony Wilson on there? 8)

(Tannin=Tony Wilson)

 

[skeet]

No relation, I can assure you (although I often find it confusing myself).

Post Republic's T.Wilson has a degree in Russian from Oxford University. Now he really does correct your spelling, syntax, grammar and Americanisation’s. He has also got a stunning German girlfriend called Lima, so I can assure you they are not the same person. Imaginary friends are nice but they are up there with vapour-ware when the lights go out at night (nudge nudge, Tea anyone).

Besides, Tea's name is not Tony! Surely you've not been seduced by the master of re-invention!

 

[Tea]

Besides, Tea's name is not Tony!

And I zhould hope not!

 

[skeet]

Sorry about that Tea, my best wishes to your partner.

I hope this is not impertinent, but are you aware that there are charity organisations in Oz that can assist extrication from arranged marriages.

 

[Cliptin]

Sorry about that Tea, my best wishes to your partner.

I hope this is not impertinent, but are you aware that there are charity organisations in Oz that can assist extrication from arranged marriages.

Eewwww! Ballarat is not that rural.

 

[Tea]

I'm much too young to be married. And if I wasn't, what on earth makes you think I'd marry Tannin? (I think P5 is more my style. At least he has a proper amount of fur. Or possibly The Giver, because he liked my poem.)

 

[skeet]

Yes of course Tea, silly me. Ape and cow is never really going to work is it. My humblest apologise and please accept this banana as a token of my genuine remorse.

I will also withdraw gracefully from this most esteemed forum and would like to introduce a friend of mine, Ekee Ekee Ekee Fatang Fantang ala Biscuit Barrell.