Originally thought that I was posting new information last night (I was so excited to share about the diskless version), but after seeing an earlier post from Rob53, I realize I was being redundant and not adding anything new.
Rewrote my post to more confirm what Rob53 wrote, with my personal observation. I use my Pegasus2 R4 to hold my Lightroom Catalogues. So far I'm very pleased with the performance.
I purchased and received an diskless R4 case from Apple directly in late December and put my own drives into it. The process was fairly easy, but the instructions didn't really consider the diskless version.
There is a mixture of two manufacturers (all 3TB drives) in the array, and although it took ALL night to create the RAID6 drive, by my second cup of coffee in the morning it was ready. So far it works perfectly, and it has the capacity and speed that I would expect to have. The apple forums have people who claim to have put 4 TB drives inside, but I can't confirm, and of course they are rather expensive at the moment.
Do you have to install a driver under OS X for this device to be recognized as a single drive?
Does the RAID level in this device manage itself once it has been built without relying on OS X? I realize in order to initialize and build a RAID level, this device has to be connected to a Mac running OS X.
This is a RAID cabinet so it has the hardware to handle the array management built-in. Once it's configured, it should present it's self as one storage device to whatever it's plugged into. Technically, it looks like a raw device to whatever it's plugged into [theoretically, it could have multiple LUNs and present it's self as multiple raw devices]. If it's got a file system on it.. it'll mount. If it doesn't it'll prompt you to partition/format it.
If it didn't have the RAID hardware built in, it would be considered a JBOD [Just a Bunch Of Disks]. JBODs will show up as individual disks when the device is plugged in and they can only be setup as a RAID if they're plugged into a RAID controller or if the computer creates the RAID with software. That arrangement wouldn't be as portable.
It'd be nice to see some bandwidth tests in RAID0 or with the included drives swapped for SSDs. From the numbers I'm seeing, this is what I'd expect from a small array of 7200RPM SATA drives.
I'm not sure if the peak numbers reflect read and write from the cache.. not sure what you were doing to generate I/O. I'd hope the cache is faster then 500Gb/sec.
I'm assuming the R6 and R8 use essentially the same controller if not an identical controller. If we knew what the upper limits the controller on TB2 were, we'd be able to make some informed guesses about how performance might scale to the larger cases.
Phil, not sure if you mean what you said but you always want to use matched drives in a RAID. This way they are all spinning the same and work the same, achieving the best performance. Promise sells most of their RAID systems with drives, not drive-less (like Areca). Apple sells a diskless version meant to be used with the old Mac Pro's 4 disks, http://store.apple.com/us/product/HE150VC/A/promise-pegasus2-r4-diskless-4bay-thunderbolt-2-raid-system?fnode=5f but everything else comes with matched drives.
Excellent! Thanks for the link.
Well, maybe the make/model makes sense with your statement, but wouldn't you agree that getting drives from different batches will be safer than getting them from a single batch? There are a few data centers experts here, and more than one have stated that different batches is safer.
Well, maybe the make/model makes sense with your statement, but wouldn't you agree that getting drives from different batches will be safer than getting them from a single batch? There are a few data centers experts here, and more than one have stated that different batches is safer.
I understand the desire to have matching drives on the front end. But I think we should expect a RAID system to handle non-matching drives fairly well, as long as we conform to some standards. Even if we go to some trouble to match drives when we create our drive, eventually one or more drives will fail and need to be replaced. Should we expect that replacement drives be matched as well? I think this can be a slippery slope.
The OS running on a computer to which this RAID box is connected has to have a driver if this is a hardware RAID. I bet Apple has included the driver in OS X for this RAID DAS.
My goal is to connect such a RAID DAS to VMware ESXi. It would have to have the correct driver installed to recognize the hardware RAID controller in the Pegasus.
Well, maybe the make/model makes sense with your statement, but wouldn't you agree that getting drives from different batches will be safer than getting them from a single batch? There are a few data centers experts here, and more than one have stated that different batches is safer.
Definitely use the same Make and Model. You want the members of the Array to all have the same performance characteristics. When you're taking a stream of data, breaking it up, and writing it across multiple disks, you don't want to have 1 member with slower write speeds than the others.
You can make an argument either way on batch.
Your co-workers are taking the position that the drive manufacturer may have production runs that have common manufacturing issues. As a result, if you just happen to have 4 drives from that run in your Promise box, you're likelihood of a multi-drive failure is higher. Single failures are a pain.. multiple contemporaneous failures are a disaster.
It's not an unreasonable assumption.. I've seen more than a few examples of bad batches of components over the years.. everything from iBook video connectors to video chips to hard drives.
We had a 36hr work day last year when multiple Seagates in one of our SAN boxes died before the first and second spare could rebuild parity. Given we were running RAID 50 with two spares.. uptime was a big deal.
I could make a valid against mixing batches too though.. at least that it's not worth the effort.
If we accept the assumption that certain production runs of generally good drives have unusually high failure rates due to manufacturing issues during that run, then we're increasing our chance of getting a drive from a bad production run by 4x if we use drives from different production runs. We're less likely to get 4 duds but more likely to get at least one dud. Depending on how many production runs are bad versus good runs.. we might have a higher chance of getting 2 out of 4 duds by mixing.
Again.. one dead drive bad.. many dead drives unacceptable. Here's the thing though. From my experience.. about 16 years now.. I've found that drive models tend to be problematic more than specific production runs. For example.. Seagate Barracudas from the past few years are just bad drives. It's not that one out of a dozen production runs are bad.. they're all bad drives. That's even more true with the WD Greens. We bought a lot of them when they came out.. I don't think any of them are still running.. and some were replaced on warranty and those also failed.
That all doesn't necessarily argue against mixing batches as much as it argues that it's not a placebo.
A better argument mixing is that.. for the same reasons you want to have identical model drives in the RAID, you want to have the same firmware on the drives.
In reality.. the truth is probably somewhere in between..
Use the same model drives in a RAID. It probably doesn't hurt to get them from different production batches [mix them if you've purchased several batches of drives] but don't consider that a solution to multi-drive failure. Update them all to the same [current] firmware when possible before building your array.
This is opinion.. but I'd say keeping a hot-spare in the device [an idle drive not integrated in to the active array] and keeping a spare replacement in the drawer are far better policy than going through the trouble of trying to source a bunch of identical drives from different production runs.
Definitely use the same Make and Model. You want the members of the Array to all have the same performance characteristics. When you're taking a stream of data, breaking it up, and writing it across multiple disks, you don't want to have 1 member with slower write speeds than the others.
You can make an argument either way on batch.
Thank you for your elaborative post and expertise in this area. I appreciate your insight and will remember this post when replacing disks in my MP.
Getting ready to buy this unit in a day or two. How is it after you posted this ? I tried to find the info about the command line that turned off the spin-up problem but could not. Can you post the link or command please.
Was the heat problem ever fixed ? Any other tricks or tips ? Would you buy it again ?
I see B&H is offering a mini Pegasus for free when buying. Two 128meg SSD drives in a compact Thunderbolt2 case. (Retail's for approx $400.00)
Would love to hear your pro's and con's on drive before pulling the trigger.
Comments
Rewrote my post to more confirm what Rob53 wrote, with my personal observation. I use my Pegasus2 R4 to hold my Lightroom Catalogues. So far I'm very pleased with the performance.
I purchased and received an diskless R4 case from Apple directly in late December and put my own drives into it. The process was fairly easy, but the instructions didn't really consider the diskless version.
There is a mixture of two manufacturers (all 3TB drives) in the array, and although it took ALL night to create the RAID6 drive, by my second cup of coffee in the morning it was ready. So far it works perfectly, and it has the capacity and speed that I would expect to have. The apple forums have people who claim to have put 4 TB drives inside, but I can't confirm, and of course they are rather expensive at the moment.
http://store.apple.com/xc/product/HE150VC/A
Do you have to install a driver under OS X for this device to be recognized as a single drive?
Does the RAID level in this device manage itself once it has been built without relying on OS X? I realize in order to initialize and build a RAID level, this device has to be connected to a Mac running OS X.
This is a RAID cabinet so it has the hardware to handle the array management built-in. Once it's configured, it should present it's self as one storage device to whatever it's plugged into. Technically, it looks like a raw device to whatever it's plugged into [theoretically, it could have multiple LUNs and present it's self as multiple raw devices]. If it's got a file system on it.. it'll mount. If it doesn't it'll prompt you to partition/format it.
If it didn't have the RAID hardware built in, it would be considered a JBOD [Just a Bunch Of Disks]. JBODs will show up as individual disks when the device is plugged in and they can only be setup as a RAID if they're plugged into a RAID controller or if the computer creates the RAID with software. That arrangement wouldn't be as portable.
It'd be nice to see some bandwidth tests in RAID0 or with the included drives swapped for SSDs. From the numbers I'm seeing, this is what I'd expect from a small array of 7200RPM SATA drives.
I'm not sure if the peak numbers reflect read and write from the cache.. not sure what you were doing to generate I/O. I'd hope the cache is faster then 500Gb/sec.
I'm assuming the R6 and R8 use essentially the same controller if not an identical controller. If we knew what the upper limits the controller on TB2 were, we'd be able to make some informed guesses about how performance might scale to the larger cases.
Excellent! Thanks for the link.
Well, maybe the make/model makes sense with your statement, but wouldn't you agree that getting drives from different batches will be safer than getting them from a single batch? There are a few data centers experts here, and more than one have stated that different batches is safer.
Makes sense to me.
Meanwhile I dream of 5 TB SSDs for $120.
Rob,
GoneNuts says he was able to set up a heterogenous RAID, although I agree with your point about matching drives in a RAID.
But my question was, can you mix both RAID (eg 2 RAID-0 drives) and non RAID drives in the Pegasus?
Just a thought about matching drives.
I understand the desire to have matching drives on the front end. But I think we should expect a RAID system to handle non-matching drives fairly well, as long as we conform to some standards. Even if we go to some trouble to match drives when we create our drive, eventually one or more drives will fail and need to be replaced. Should we expect that replacement drives be matched as well? I think this can be a slippery slope.
We're gonna so laugh at this in
202x201x.My goal is to connect such a RAID DAS to VMware ESXi. It would have to have the correct driver installed to recognize the hardware RAID controller in the Pegasus.
Definitely use the same Make and Model. You want the members of the Array to all have the same performance characteristics. When you're taking a stream of data, breaking it up, and writing it across multiple disks, you don't want to have 1 member with slower write speeds than the others.
You can make an argument either way on batch.
Your co-workers are taking the position that the drive manufacturer may have production runs that have common manufacturing issues. As a result, if you just happen to have 4 drives from that run in your Promise box, you're likelihood of a multi-drive failure is higher. Single failures are a pain.. multiple contemporaneous failures are a disaster.
It's not an unreasonable assumption.. I've seen more than a few examples of bad batches of components over the years.. everything from iBook video connectors to video chips to hard drives.
We had a 36hr work day last year when multiple Seagates in one of our SAN boxes died before the first and second spare could rebuild parity. Given we were running RAID 50 with two spares.. uptime was a big deal.
I could make a valid against mixing batches too though.. at least that it's not worth the effort.
If we accept the assumption that certain production runs of generally good drives have unusually high failure rates due to manufacturing issues during that run, then we're increasing our chance of getting a drive from a bad production run by 4x if we use drives from different production runs. We're less likely to get 4 duds but more likely to get at least one dud. Depending on how many production runs are bad versus good runs.. we might have a higher chance of getting 2 out of 4 duds by mixing.
Again.. one dead drive bad.. many dead drives unacceptable. Here's the thing though. From my experience.. about 16 years now.. I've found that drive models tend to be problematic more than specific production runs. For example.. Seagate Barracudas from the past few years are just bad drives. It's not that one out of a dozen production runs are bad.. they're all bad drives. That's even more true with the WD Greens. We bought a lot of them when they came out.. I don't think any of them are still running.. and some were replaced on warranty and those also failed.
That all doesn't necessarily argue against mixing batches as much as it argues that it's not a placebo.
A better argument mixing is that.. for the same reasons you want to have identical model drives in the RAID, you want to have the same firmware on the drives.
In reality.. the truth is probably somewhere in between..
Use the same model drives in a RAID. It probably doesn't hurt to get them from different production batches [mix them if you've purchased several batches of drives] but don't consider that a solution to multi-drive failure. Update them all to the same [current] firmware when possible before building your array.
This is opinion.. but I'd say keeping a hot-spare in the device [an idle drive not integrated in to the active array] and keeping a spare replacement in the drawer are far better policy than going through the trouble of trying to source a bunch of identical drives from different production runs.
Thank you for your elaborative post and expertise in this area. I appreciate your insight and will remember this post when replacing disks in my MP.
Getting ready to buy this unit in a day or two. How is it after you posted this ? I tried to find the info about the command line that turned off the spin-up problem but could not. Can you post the link or command please.
Was the heat problem ever fixed ? Any other tricks or tips ? Would you buy it again ?
I see B&H is offering a mini Pegasus for free when buying. Two 128meg SSD drives in a compact Thunderbolt2 case. (Retail's for approx $400.00)
Would love to hear your pro's and con's on drive before pulling the trigger.
TIA