Luminous Landscape Forum
Equipment & Techniques => Computers & Peripherals => Topic started by: ddk on January 06, 2009, 08:29:47 pm
-
I know there's been some recent excitement regarding Drobo units, I'm writing this as warning to both new owners who're going to trust they're precious data to one or those about to buy a Drobo. One has to put with very slow write speeds compared to and a ridiculous annual service fee for firmware upgrades unlike any other RAID from a decent vendor and in return one gets the convenience of upgrading/replacing drives as needed and full security of your data; right? Right...
What the Drobo guys don't mention besides the hidden mandatory support fees is that your data isn't protected while you replace that drive for a very long time. Today I replaced one of its with a larger one and according to Drobo's software the whole upgrade process will take about 20 hrs during which time I can lose all my data in case of a single drive failure. I only have 630 gb of data and it takes Drobo 20 hrs to process, from what I read here many are setting there's up with TBs of drive space, I wonder how long its going take your Drobo to process that much data, whatever it is its going to be way too long for a drive to go south during the upgrade process and unlike a typical mirrored Raid that one can still recover data from a failed drive, its not the case with Drobo.
I don't know about you guys, if you're wiling to risk your precious data but I'm from the Murphy school of thought; "What can go wrong WILL go wrong!"
(http://www.pbase.com/ddk/image/107880222/original.jpg)
-
Um, what?
This is true of a lot of raids. While the rebuild is going on you're in a degraded state. And if the rebuild is courtesy raid 5 pairity data it can take quite a while to reconstruct that.
Even in our SANs that used to be a PITA.
-
The thing to do is use Drobo or any other RAID unit as Backup, not storage I have a Windows Home Server [I thought it was a much better product than the Drobo for many reasons] and the WHS backs up data I have elsewhere anyway, so if I need to add/change drives The data being rejigged is not just on there anyway.
-
Um, what?
This is true of a lot of raids. While the rebuild is going on you're in a degraded state. And if the rebuild is courtesy raid 5 pairity data it can take quite a while to reconstruct that.
Even in our SANs that used to be a PITA.
It never takes 20+hrs to rebuild 630 gb of data in any configuration, besides with a Raid system you have other options besides Raid 5, Raid 2 is fast to rebuild and specially Raid 6 is much more secure. You have no options with Drobo, its very slow and costs similar to the Raid solutions.
-
The thing to do is use Drobo or any other RAID unit as Backup, not storage I have a Windows Home Server [I thought it was a much better product than the Drobo for many reasons] and the WHS backs up data I have elsewhere anyway, so if I need to add/change drives The data being rejigged is not just on there anyway.
Well, Raid systems are used for secure storage and speed, Drobo is supposed to be a Raid alternative at least security wise since it can't compete speed wise with similarly priced raids. My point is that it can't be trusted anymore than a single disk drive for important backups. What it really is, is an expensive depository for your older drives.
-
Well, Raid systems are used for secure storage and speed, Drobo is supposed to be a Raid alternative at least security wise since it can't compete speed wise with similarly priced raids. My point is that it can't be trusted anymore than a single disk drive for important backups. What it really is, is an expensive depository for your older drives.
Any system that has a vulnerable time whilst being rebuilt should not be the only storage was my point. Two completely separate storage devices are better even if one or more of them happen to be RAID or similar as well.
In this case, it would be better to have your data on two Drobo boxes, so one would always be safe, if the other has down time to rebuild when changing/replacing drives.
Which reminds me. It's time to buy more hard drives. Again.
I'll be able to build a walli n the back yard soon out of old HDs.
-
Of course you can use the Drobo as primary storage in the same way as you can use any other RAID system. you just HAVE TO BACK UP or DUPLICATE reliably your primary storage. That's true for any kind of storage system regardless of their particular RAID, fault-tolerance or rebuild performance characteristics.
No single RAID system is a fail safe system without appropriate backup or data duplication / replication.
-
I think that, given two years from now, when the price of SSDs is WAY down, Raid will be a thing of the past. One 2.5" drive will run both your notebook and desktop, and your backup. Why?
Size!
Your 2.5" notebook drive in SSD format can easily carry 5 terabytes...anyone looking at a 32G Sandisk Flashcard right next to a 2.5" drive can easily assume this.
Rotating platters are the only thing today that are still "dating" our computers. The only physically moving parts on a CPU should be the keyboard keys and the lid.
According to Laptop Magazine, CPUs have increased their performance capacity 1000-fold while hard drive access times have increased by less than twice. I forget from what year this references.
-
I think that, given two years from now, when the price of SSDs is WAY down, Raid will be a thing of the past.
Why? Do SSD come with a 100% lifetime guarrantee against failure and loss of data? No? I didn't think so...
-
Why? Do SSD come with a 100% lifetime guarrantee against failure and loss of data? No? I didn't think so...
I seem to vaguely recall reading about problems with SSDs, now they have become common enough for people to discover real world issues.
-
Of course you can use the Drobo as primary storage in the same way as you can use any other RAID system. you just HAVE TO BACK UP or DUPLICATE reliably your primary storage.
My post should have read 'only storage' not primary/only storage'. Oops It's corrected now.
-
No single RAID system is a fail safe system without appropriate backup or data duplication / replication.
Indeed. This should be answer number one for every RAID FAQ. Anybody relying on a single RAID 5, 6, or 7 solution for critical data storage is asking for trouble.
-
Indeed. This should be answer number one for every RAID FAQ. Anybody relying on a single RAID 5, 6, or 7 solution for critical data storage is asking for trouble.
Yep. We used to see a lot of drives killed by the rebuild. You'd lose one and then the strain of use plus rebuild would be too much for the next weakest drive.
Beyond that, however, is the possibility of a chassis failure. All your drives can be fine but if the raid unit itself pines for the fjords you're sol.
-
Yep. We used to see a lot of drives killed by the rebuild. You'd lose one and then the strain of use plus rebuild would be too much for the next weakest drive.
Beyond that, however, is the possibility of a chassis failure. All your drives can be fine but if the raid unit itself pines for the fjords you're sol.
Or there's a fire, flood, theft, etc. RAID is not meant for backup, but for uptime and/or speed.
-
BTW, anybody looking to *sell* their gen2 FW800 Drobo, I am in the market for a second one!
-
Why? Do SSD come with a 100% lifetime guarrantee against failure and loss of data? No? I didn't think so...
Do raid come with a 100% lifetime guarrantee against failure and loss of data? No? I didn't think so...
With their MTBF (mean time before failures) being much much less than that found on rotating platter drives, what would you rather risk? Five slow rotating platters or one SSD with 20% of the chance of going bad as a single drive?
Also, you can always RAID SSDs. One's performance read/writes under SSDs, I am saying, will practically make these exclusive. No one but Google and large data centers will be using old-school hard drives ten years from now.
-
Do raid come with a 100% lifetime guarrantee against failure and loss of data? No? I didn't think so...
With their MTBF (mean time before failures) being much much less than that found on rotating platter drives, what would you rather risk? Five slow rotating platters or one SSD with 20% of the chance of going bad as a single drive?
Also, you can always RAID SSDs. One's performance read/writes under SSDs, I am saying, will practically make these exclusive. No one but Google and large data centers will be using old-school hard drives ten years from now.
1st: I suppose you got your MTBFs the wrong way around otherwise your post doesn't make sense!
2nd: I would rather take no risk
3rd: You said RAID was dead, to which I gave my answer, now you tell me RAID isn't dead? Glad you came to your senses!
-
Roger.
But I think that in five years' time you won't be using RAID; nor will most people on this board.
-
Drobo's early advertisements boasted that you will "never" lose an image again! Sounds like a lawsuit waiting to happen.
I have a friend who owns 3 Apple retail stores. He sold Drobos but had so many problems with the units and the lack of support from the company, that he stopped selling them. The early USB units had reliability issues. I'm not sure how they are now, but the concept of their "proprietary black box" scares me.
I'll stick with what I know works well - redundant JBOD towers.
My 2 cents.
Bud James
-
But I think that in five years' time you won't be using RAID; nor will most people on this board.
Since devices have a tendency to fail once in a while, RAID (redundant array of inexpensive disks) will be a popular option among those of us who care for our data.
What "most people on this board" do now or in five years is pure speculation, though.
Sometimes there are other parts than the actual storage bits of the drive that fail, such as the control logic, and it would be strange if that couldn't happen to an SSD with approximately the same regularity as in HDDs.
[attachment=10891:IMG_4547.jpg]
[attachment=10890:IMG_4545.jpg]
Then above is an image of a controller chip that lost its magic smoke (you know, the smoke that's inside all electronic components, and when let out leads to immediate failure ). If that disk hadn't been in a RAID, then the data might have been unrecoverable.
Fortunately, harddisk controller cards are actually user serviceable (if you know what you're doing), so I managed to save a long rebuild time by getting the card from a faulty disk with a working card. I haven't seen SSDs which appear to be anywhere near as easily fixed.
You are of course free to gamble that your data will be just as safe without RAID as with. I won't. And I'll still be making backups.
-
I have a friend that setup a far better system IMHO for far less money. He purchased four 750 gig super fast drives. Then he built them, using the Apple supplied Raid software (disk utility) as mirrored and striped array! So four drives, two mirroring for protection while striping for speed. Cost about $500 (popped them in his existing four drive bays in his MacPro).
-
I have a friend that setup a far better system IMHO for far less money. He purchased four 750 gig super fast drives. Then he built them, using the Apple supplied Raid software (disk utility) as mirrored and striped array! So four drives, two mirroring for protection while striping for speed. Cost about $500 (popped them in his existing four drive bays in his MacPro).
It seems unlikely that he purchased "super fast" drives, as the only 750 GB drives available only spin at 7200 RPM (or, in the case of Western Digital's Caviar Green line, varying between around 5400 and 7200 RPM).
But Seagate has just announced that they will be selling 800 GB Cheetah drives (spinning at 15000 RPM), which I suppose is necessary to keep ahead of the SDD competition for a few months.
Your friend's setup, if he did it right, is called "RAID 10", and is one of the best combinations of performance and reliability you can get in four drives.
Up to two drives can fail without loss of data, but if the second one's the wrong disk, you're of course not on the sunny side of luck anymore.
-
I have a friend that setup a far better system IMHO for far less money. He purchased four 750 gig super fast drives. Then he built them, using the Apple supplied Raid software (disk utility) as mirrored and striped array! So four drives, two mirroring for protection while striping for speed. Cost about $500 (popped them in his existing four drive bays in his MacPro).
While we're on the subject, this is the set-up I use for performance when reading, batch processing and writing large blocks of files. I am using 4x WD 640G Caviar non-enterprise drives*. The first two are striped for the OS and and the second pair are striped with a thin outer 128 G partition for scratch and the rest for my working image files. With this set up, I can read from the working file array and write to the OS desktop and get sustained transfer speeds of 150MB/s, which is pretty decent -- C1 literally screams as I batch convert a block of my P45+ raws. And yes, having my CS scratch on a stripe really improves performance when manipulating large files that need scratch. Also, a FWIW benchmark: CS4 opens in less than 3 seconds from the striped OS array.
* At present, the WD 640's are among the fastest currently available 7200RPM SATA2 non-enterprise drives THAT RAID WELL. Hitachi 1TB non-enterprise spinpoints benchmark very fast for 7200 RPM drives too, but myself and others have reported problems with them in RAID configurations. And the WD 640's are dirt cheap now.
But that isn't reliability, it is performance... And I recommend you have a solid back-up strategy if your main files are on raid arrays or single drives. My current strategy is I back up my working image files to the Drobo (which also stores my historic image files) and then back that up to individual drives stored offsite. A bootable copy of my OS is backed up to a partition on a 5th drive stuffed into the lower optical bay of the MacPro box and connected to one of the extra SATA ports on the MB -- that operation is scheduled weekly using Carbon Copy Cloner, and yes I've needed it. This drive also carries a Time Machine partition which I've rescheduled to run twice daily (instead of hourly) simply to serve as tertiary back up of the OS, documents, mail and downloads.
-
"Your friend's setup, if he did it right, is called "RAID 10"..."
Well, maybe that's becoming a common shorthand, but more correctly it's RAID 0 + 1. I too built one years ago with a damned expensive SCSI system. Turned out to be even more expensive keeping it running, and it's long gone.
Currently I'm using the very fast WD Raptor drive... http://www.storagereview.com/WD3000BLFS.sr (http://www.storagereview.com/WD3000BLFS.sr) ...the 150, for data. I'm aggressive at culling trash*, so it's plenty big enough.
* A very active fellow fotog, up to a year or so ago, had sort of never tossed any pic taken--he has over a dozen external drives connected to his machine, and NONE are backed up!
-
One more thing, boys and girls. Do understand that there are only 2 kinds of harddrives--those that have failed and those that will.
Backing up your digital pics is just like flossing your teeth--you need to backup or floss only those you wish to keep.
-
"Your friend's setup, if he did it right, is called "RAID 10"..."
Well, maybe that's becoming a common shorthand, but more correctly it's RAID 0 + 1.
No, "RAID 10" is shorthand for "RAID 1+0". RAID 0+1 has the shorthand "RAID 01".
These are fundamentally different, and I strongly advise people not to use RAID 0+1 because of the reduced reliability compared to RAID 1+0 and RAID 6.
RAID 0+1: mirror of stripes (you can safely lose up to N drives, but only if they're in the same striped set)
RAID 1+0: stripe of mirrors (you can safely lose any drive in any of the N mirrors, that is, up to N drives)
Typically, N = 2 (four disks).
-
Hitachi 1TB non-enterprise spinpoints benchmark very fast for 7200 RPM drives too, but myself and others have reported problems with them in RAID configurations.
I can attest to that, at least anecdotally. (I don't work for e.g. NetApp or EMC, so large quantity statistics are not available to me.)
At work, we used to use 500 GB and 1 TB Hitachi drives because these were the first to offer those capacities (where we are).
These caused us nothing but grief, in RAID 1+0 or RAID 5.
Changing to Seagate and Western Digital drives lessened our pain significantly.
Changing to SAS drives has so far removed the pain, and our customers are far less dissatisfied.
(Happy customers rarely speak up, it's the dissatisfied customers who get in touch.)
-
While we're on the subject, this is the set-up I use for performance when reading, batch processing and writing large blocks of files. I am using 4x WD 640G Caviar non-enterprise drives*. The first two are striped for the OS and and the second pair are striped with a thin outer 128 G partition for scratch and the rest for my working image files. With this set up, I can read from the working file array and write to the OS desktop and get sustained transfer speeds of 150MB/s, which is pretty decent -- C1 literally screams as I batch convert a block of my P45+ raws. And yes, having my CS scratch on a stripe really improves performance when manipulating large files that need scratch. Also, a FWIW benchmark: CS4 opens in less than 3 seconds from the striped OS array.
* At present, the WD 640's are among the fastest currently available 7200RPM SATA2 non-enterprise drives THAT RAID WELL. Hitachi 1TB non-enterprise spinpoints benchmark very fast for 7200 RPM drives too, but myself and others have reported problems with them in RAID configurations. And the WD 640's are dirt cheap now.
But that isn't reliability, it is performance... And I recommend you have a solid back-up strategy if your main files are on raid arrays or single drives. My current strategy is I back up my working image files to the Drobo (which also stores my historic image files) and then back that up to individual drives stored offsite. A bootable copy of my OS is backed up to a partition on a 5th drive stuffed into the lower optical bay of the MacPro box and connected to one of the extra SATA ports on the MB -- that operation is scheduled weekly using Carbon Copy Cloner, and yes I've needed it. This drive also carries a Time Machine partition which I've rescheduled to run twice daily (instead of hourly) simply to serve as tertiary back up of the OS, documents, mail and downloads.
I love your setup and am looking to upgrade to a MacPro and take advantage of having four drive bays over my current laptop which has to be backed up to my server. I think the biggest advantage is just being able to swap out drives as they fill up. The Drobo is a great unit, but I worry about the combination of the proprietary technology and the single point of failure in the HW RAID card/software. Being able to easily access my data 20 years from now is a priority and with a simple HD all that's needed is a SATA interface. I have no idea how one would access data off a Drobo pulled out of storage that had a HW failure and the company is no longer around or no longer supports the particular model. I'm probably paranoid, but figure my clients appreciate that.
-
I use a RAID as my primary storage. It's really more of a SAN, but it's 2003 server with an Areca 1230(12 port) card in it.
Critical data is run on Raid 6 Partitions (3 drives must fail to lose data) while non-critical data is stored on Raid 5 (2 drives must fail to lose data.)
I have an open slot on the system where I slide in backup drives and copy data over at regular intervals using robocopy. I'm sure it would be better if I used backup software with compression, but I want to be able to have quick access to specific files if the system fails and I can plug those backup drives into any computer to recover the data.
Eventually when Bluray comes down in price per disc, I might consider old methods of backup and copy to 50gb bluray, but not at current prices.
-
A few years ago I had a surge that wiped out the memory and hard drives on my G5 desktop . But only the harddrives in the machine my externals were fine. I will never make copies off the hard drives in my computer. They are all to be copied externally.