RAID in Early 2009 Mac Pro

[mikey1004]

Being the "tech guy" of the house, my brother has requested that I upgrade his 4,1 Mac Pro. He currently has:

2 x 2.26 Quad-Core Xenons
12 GB RAM
3 x GeForce GT 120
4 x 2 TB HDD's in a software RAID 10
80 GB SSD in one of the optical bays as a boot disk.

He primarily uses his computer for video editing, mostly in Final Cut Pro X, but sometimes he'll dabble in the Adobe suite. So he said that he wants more speed and needs less storage; i.e., he doesn't need his main drive to be more than 2 TB. Budget is not really an issue here, but future-proofing is: I don't want to have to upgrade anything for a long time.

So RAM upgrade is simple enough, right? Just head over to OWC and grab 24 or 32 GB of whatever they say goes with the 4,1 generation Mac Pro.

Video cards I know next to nothing about. I don't quite understand why he's got three of them for; his only monitor is a 30 in. cinema display. Should I upgrade his video card(s)? Maybe the Raedon 5770 or 5870? I feel like video editing isn't really all that GPU-intensive, but if it is, please tell me so.

Now, the fun part is the storage. I figure the best solution is get four new SSD's to replace the 12 TB ones, and stick the SSDs in a RAID 0. So the question is: do I need a RAID card for that? Or, I guess, more accurately: do I want a RAID card for that? And if I do, which one? I assume not the Apple one, as I see it's not well liked.

Then I guess I could get an external RAID enclosure thingie for the leftover disks and use them for backup/additional storage.

Also, do I want to upgrade the CPU? I still feel like the bottle neck for him is not in the CPU but in read/write times, but if what he has is really outdated at this point (or will be in the near future), maybe I should fix that.

 

[Inconsequential]

If he ever pegs all 8 cores (not necessarily all 16 threads), then CPU upgrades (especially to 6 core chips) would be a HUGE boost.

He may have 3 GPUs for some sort of GPU acceleration, but I don't think the GT120 can do that?

GPU upgrade in terms of perf: 5870 (Apple) > 6870 (Flashed) > 5770 > 4870

If he has page-outs, then more memory.

Surely a 256GB/512GB SSD (270MB/sec read and write) should suffice?

If not he would need a SATA 3 PCI-E card to get the full 500+MB/sec read and write.

You are limited to 660MB/sec due to the chipset and 270MB/sec via SATA 2, so you WILL need a PCI-E 4x card or better to beat that.

So perhaps two Intel 320 300GB drives in RAID 0 and then two WD Black 1TB for data?

I don't think he needs more than 300MB/sec tbh.

 

[wonderspark]

Yeah, the 3x standard GT 120 GPUs for one Apple monitor is odd to me. Does he want more monitors? I have a 5870 for a 22" Dell and a 30" ACD, and it's great. If he goes beyond "dabbling" in Adobe, he may want a card that takes advantage of CUDA, but personally, I blew that off and just use software acceleration only in CS5 / 5.5 and have no problems.

32GB of RAM should stop page outs and swap file use. It did for me, and it's cheap.

I also find it strange that he indicated "less storage," but then again I need a ton of storage for multiple projects at once. He has 4TB now with his RAID 10 setup, so if all he wants is more speed, you could simply set up three of the 2TB disks in RAID 0 and use the fourth one for backup without any need for a RAID card. He'd have over 300MB/sec that way (I got 330MB/sec with an internal RAID 0 of 3x2TB disks), and since he's not using much of the space, he'd be using the fastest parts of the disks in that RAID. So long as he doesn't use more than 2TB, that fourth disk will be great as a backup.

I'm just trying to save you all money by putting what you have to better use, based on how he seems to be using it.

I have a RAID that I dumped some money into, but it's pretty fantastic. It's an 8-bay external box (by Sans Digital) filled with 8x Western Digital 2TB RE-4 disks in RAID 6 via an Areca 1880ix-12 card. It gives me 12TB of storage with the safety of allowing any two of the eight disks to fail without losing any data. It provides read speeds of 714MB/sec and write speeds of 816MB/sec based on my own tests, and I can even edit native, unrendered DSLR footage while rebuilding the RAID during a disk failure. (Another test I did by pulling a drive while editing.) You might be able to build something with similar speeds using SSDs, but the storage capacity will not come close, however he doesn't seem to need capacity so it's an idea. I'm personally not ready to trust an SSD RAID just yet, and my livelihood depends on my system.

I think the 2.26GHz CPUs are slow, and if you're brave enough to swap in new CPUs, it can be done with some specific sized washers to compensate for the difference in height between lidded and un-lidded CPUs. It would make a huge difference.

 

[deconstruct60]

Being the "tech guy" of the house, my brother has requested that I upgrade his 4,1 Mac Pro. He currently has:

2 x 2.26 Quad-Core Xenons
12 GB RAM
3 x GeForce GT 120
4 x 2 TB HDD's in a software RAID 10
80 GB SSD in one of the optical bays as a boot disk.

He primarily uses his computer for video editing, mostly in Final Cut Pro X, but sometimes he'll dabble in the Adobe suite. ...

So RAM upgrade is simple enough, right? Just head over to OWC and grab 24 or 32 GB of whatever they say goes with the 4,1 generation Mac Pro.

Video cards I know next to nothing about. I don't quite understand why he's got three of them for; his only monitor is a 30 in. cinema display. Should I upgrade his video card(s)? Maybe the Raedon 5770 or 5870? I feel like video editing isn't really all that GPU-intensive, but if it is, please tell me so.

Not sure this is a RAID problem.

1. FCPX can use GPGPU to offload some computations. So yes it can be intensive. Adobe Premiere even more so (but will need Nvidia cards to fully leverage that).

2. What are the 3 GT120s driving? It may be better to go to a two 5770 set-up than to trying to absorb all of the video driving into one card. ( again this goes back to the GPGPU workload. If you saddle the card with driving three monitors it is not going to have as much resources to do GPGPU work.
Two cards where one handles some lightweight monitor (2D only work; email/chat/etc ) and GPGPU work and the other larger monitors may work better.

( again if the Adobe Primiere is an important factor two cards fit the 5770 power profile maybe a better bet. )


3. If bump the memory and graphics that should be enough for step update to the configuration. He can use those for a while to get a new baseline of what the performance bottlenecks are and then spend some more incrementally to remove the largest, most limiting, more next.

Changing too many things at a time only somewhat invites trouble. if there isn't going to be a "back up" machine to fall back to. It is another issue if buying a new box and can fall back to the older one if there is a migration/upgrade problem.

Now, the fun part is the storage. I figure the best solution is get four new SSD's to replace the 12 TB ones, and stick the SSDs in a RAID 0.

4 SSDs for a 2009's internal controller is bad idea. Well perhaps not bad, but won't be a balanced system. 4 SSDs in RAID 0 isn't good either. For the target major application, video, it is likely better to just keep the RAID 10 HDD set-up. You can look to see if there are newer HDDs with better sequential performance that what is in there if the drives are 2+ years old.

So the question is: do I need a RAID card for that? Or, I guess, more accurately: do I want a RAID card for that? And if I do, which one? I assume not the Apple one, as I see it's not well liked.

First, need to settle whether need/want two video cards before moving onto a RAID card.

I suspect some will suggest a 8x PCI-e "real RAID" card here to either bypass the internal SATA bandwidth problems or to connect to some external drive sleds ( bandwidth to back-up storage seems to be missing here so far .)

Then I guess I could get an external RAID enclosure thingie for the leftover disks and use them for backup/additional storage.

If willing to continue dealing with the software RAID 10 overhead this could be just a more straightforward x4 PCI-e SATA expander/switch card with 4 (or 8 ) ports to the external box. Or a card limited to RAID 0,1,10 but would need to back-up/restore the data since the drives would need to be reformatted.

Also, do I want to upgrade the CPU? I still feel like the bottle neck for him is not in the CPU but in read/write times,

Can be measured. Activity Monitor is a built-in tool. There are others that do more precise profiling.

but if what he has is really outdated at this point (or will be in the near future), maybe I should fix that.

If a substantial fraction of the long running computational workload switches over to the GPGPU processing then may not need faster CPUs in the short term. It is another reason to adjust incrementally as change major system components. Measure now for a baseline. Then measure after major GPU upgrade to see if open/closed the bottleneck.




P.S. There was a config that Apple sold in BTO at one point that had 3 GT120 in it. I think those were targeted mainly to those that had 3 (or more ) monitors to drive. At one point there may have been more monitors. If it is really only one monitor and will only be one monitor in the future then dropping to one GPU card makes sense.

 

[mikey1004]

Wow, you guys are fantastic!

Concorde Rules, do you know of any well regarded PCI-e RAID cards off the top of your head? If not, I'll head over to Newegg (or, you know, Google).



WonderSpark, he didn't specifically ask for less storage, but he only ever works on one project at a time, and then clears everything out before he starts a new one. So, for example, he's working on an hour-long movie right now, which has taken up about 1.3 TB.

Also, I really like your idea of just switching his RAID 10 to a 3 x 2 TB RAID 0 plus a backup disk. If he wants still wants faster, we can go the down your (more expensive) road. I assume that's the only way to do it, right? If you get a RAID controller, the disks go in an external exclosure?

This is certainly NOT his livelihood, more of a hobby. In fact, he's trying to make it in the 2D animation/cartoon world.

I didn't know there was such thing as a lidded or un-lidded CPU. Could you explain this to me in a little more detail and/or point me int he direction of something that does? And do you have any CPU's in particular that you would recommend?



deconstruct60, believe it or not, the video cards are just driving the one monitor. I have no idea why. I was not consulted before this purchase, so that would explain it, really. I double checked with him, and he does not have any desire for more displays, the 30-incher is sufficient. Premiere, as of right now, is not a primary concern; he wants the system optimized for FCPX. If optimizing for Premiere is the same thing, fantastic. If not, whatever. So you think I should go with two 5770's? Even if he only has one monitor?

I agree with your philosophy in principle: one step at a time. However, we are approaching a downtime between projects for him, so it would not be a problem if his computer were out of commission for a couple days. So this really is the time to upgrade, so I don't want to miss out on anything obvious upon which I should be improving.

So if I read your tone correctly, you don't seem to be too fond of my plan to include SSDs in his rig. So what do you think of the one proposed above?

If willing to continue dealing with the software RAID 10 overhead this could be just a more straightforward x4 PCI-e SATA expander/switch card with 4 (or 8 ) ports to the external box. Or a card limited to RAID 0,1,10 but would need to back-up/restore the data since the drives would need to be reformatted.

Here's where you lost me. (Maybe now you can see why I put "tech guy" in quotation marks.) Reformatting and restoring will be no problem, as he will have completed his projects and cleared off most of the 4 TB. What are you saying I should get?

If a substantial fraction of the long running computational workload switches over to the GPGPU processing then may not need faster CPUs in the short term.

This is also a little confusing to me. How do I figure out if the workload switches to the video card?



Also, I guess I'm a little bit lost about where all the drives would end up. If I end up sticking the the old discs, no problem. But say I opt for a RAID setup with SSDs, where do they go? An external exclosure? If so, then couldn't we leave the old discs where they are?

But still, at least for now, I think I'll go with the incrimental upgrades: RAM, GPU, switch from RAID 10 to RAID 0, and then see how that goes. If after that it turns out the CPU is the weakest link, we'll upgrade that, too.

 

[wonderspark]

Wow, you guys are fantastic!

WonderSpark, he didn't specifically ask for less storage, but he only ever works on one project at a time, and then clears everything out before he starts a new one. So, for example, he's working on an hour-long movie right now, which has taken up about 1.3 TB.

Also, I really like your idea of just switching his RAID 10 to a 3 x 2 TB RAID 0 plus a backup disk. If he wants still wants faster, we can go the down your (more expensive) road. I assume that's the only way to do it, right? If you get a RAID controller, the disks go in an external exclosure?

Not necessarily... my RAID card is an 'ix' model which means internal and external connections. (i=internal, x=external)
Not that it matters so much, as you can get cables to run internal ports externally, like I did with mine. Anyway, you could RAID three or four HDDs or even SSDs internally with a RAID card (like mine) with some specific cables for power and data, and have the full speed they provide instead of being capped by the Mac Pro's SATA II speed.

This is certainly NOT his livelihood, more of a hobby. In fact, he's trying to make it in the 2D animation/cartoon world.

I didn't know there was such thing as a lidded or un-lidded CPU. Could you explain this to me in a little more detail and/or point me int he direction of something that does? And do you have any CPU's in particular that you would recommend?

2009 dual processor Mac Pros came with lidless CPUs, meaning there is no enclosure cap on them like the ones you'd buy to upgrade them, so when you put in normal, lidded CPUs for an upgrade, you have to compensate for the difference in height with certain washers so you don't crush the new CPUs when tightening down the heat sinks.

 

[mikey1004]

Not necessarily... my RAID card is an 'ix' model which means internal and external connections. (i=internal, x=external)
Not that it matters so much, as you can get cables to run internal ports externally, like I did with mine. Anyway, you could RAID three or four HDDs or even SSDs internally with a RAID card (like mine) with some specific cables for power and data, and have the full speed they provide instead of being capped by the Mac Pro's SATA II speed.

Ah. So if you leave the SSDs on the inside, do they just go in those 4 HDD bays (with a some sort of 3.5" to 2.5" bay adapter)? Or somewhere else? Not that I plan on making this upgrade, but just for future reference in case we do end up taking this step.

2009 dual processor Mac Pros came with lidless CPUs, meaning there is no enclosure cap on them like the ones you'd buy to upgrade them, so when you put in normal, lidded CPUs for an upgrade, you have to compensate for the difference in height with certain washers so you don't crush the new CPUs when tightening down the heat sinks.

I see. I've looked through a few threads about this now, so I think I understand. Which CPU(s) would you recommend? If I understand correctly, it appears that the simplest path is sticking with a 2-chip setup, right?

 

[deconstruct60]

Also, I really like your idea of just switching his RAID 10 to a 3 x 2 TB RAID 0 plus a backup disk. If he wants still wants faster, we can go the down your (more expensive) road.

Failure of RAID 0 striped across two disks is 2x the probability of a single disk failure. Across 3 disks is 3x. Across 4 disks is 4x. Unless running extremely frequent back-ups this means you'll loose recently done work in a failure.

Stripe sets of 3 and 4 are nice for "drag racing" exercises. They aren't really prudent for individual workstations where trying to get work done.

If want something "faster" than the RAID-10 set up have now and want to throw away the failure protection (the "1" part), then switch the set-up to two RAID 0 bundles.

RAID 0 group 1 -- the sources
RAID 0 group 2 -- the outputs

So if reading and then streaming back out that is being written to two different sets of spindles. That will speed up throughput more than effectively binding 3 spindles together as a group.

Since the sources relatively static during a project it is only really the "output" set that needs frequent incremental back ups. If archive the sources in a single folder hieriarchy recovering those in a failure situation is just copying the folder back into place. ( Not very hard).

The core principle is the same as moving the OS/Apps off to a different drive (spindle). Remove that disk I/O from reading/writing large sequentially allocated files and the amount of psuedo-random I/O sent to the HDDs drops. Same thing if remove the output writes from the streaming read requests than reduce random I/O at the drive controller level.


Many Apple products tend to want to heap everything into one big pile inside of a single folder. Having to deal with data in multiple locations will actually exposure more features of FCPX than letting it do the automagic "copy and merge into blackhole" default mode.


That said RAID 0 carries risks that are best laid out for the user so they understand the trade-off they are making. It will take much longer to clean up the mess after a failure. They will likely loose what they were working on just before the failure. If reworking effort and being "down" for a day or so is OK, then it is a reasonable trade off. If not then should avoid RAID 0.
The fact the original set up is not RAID 0 implies that at some point that safety factor was a concern.

This is certainly NOT his livelihood, more of a hobby. In fact, he's trying to make it in the 2D animation/cartoon world.

If it is truly a hobby then being down for a day or two will probably work.

deconstruct60, believe it or not, the video cards are just driving the one monitor.

If it is a 30" Cinema display that's kind of hard since it has just one input. and three GT 120's are capable of driving six 30" Cinema displays. It sounds as though there are two cards doing nothing. So yes, it probably would be more optimal to reassign those PCI-e slots to something else.

If not, whatever. So you think I should go with two 5770's? Even if he only has one monitor?

I'm not sure that editing 2D animation is really going to stretch FCPX for enough to justify the second 5770 just to "unload" the main 5770 for other apps and their OpenCL processing. There are some subcomponents of FCPX that heavily leverage OpenCL and others that don't. However, since he wants to do everything on a just one monitor dual 5770's won't buy much. [ FCPX doesn't work so well when the program's windows are split over two video cards. However, can make better use of the single card if can move other GUI and/or GPGPU workload off. ]

Just one 5770 is would be a significant improvement to the the single GT 120 he is running FCPX on now. A 5870 would be a margin improvement over that. Not sure really getting "bang for the buck" with that marginal improvement; there is a $200 gap between the cards. So a 80% increase in price for often nowhere near 60-70% increase in performance. (on paper it is a 100% performance increase but that will only show up in corner cases ). Both cards are discontinued in the Windows PC market. If there "extremely tight" project deadlines it might be worth it, but as a hobby project. No. That money could be much better spent on other things (not necessarily hardware) that has a much better return on investment.

I agree with your philosophy in principle: one step at a time. However, we are approaching a downtime between projects for him, so it would not be a problem if his computer were out of commission for a couple days.

And that is the only project he is going to work on for the next 1-2 years? That project will end too. And then can do another set of upgrades. "Project end gaps" are not a good justification for even larger "Big Bang" migrations.

So if I read your tone correctly, you don't seem to be too fond of my plan to include SSDs in his rig. So what do you think of the one proposed above?

I commented on above on the RAID-0 across three disks. SSDs are a significant space/speed tradeoff. The "elbow" room he has with 4TB effective may seem "too large" but if not sinking megabucks into SSDs you'd likely turn the effective usable storage space into something less than 1TB.

Here's where you lost me. (Maybe now you can see why I put "tech guy" in quotation marks.) Reformatting and restoring will be no problem, as he will have completed his projects and cleared off most of the 4 TB. What are you saying I should get?

I'm saying you don't need a fancy, very high speed RAID card to put 4 disks in an external enclosure. There are cards which will give you eSATA or SAS/SATA external connections that don't do any "RAID". You can just get an external connection to 4 disks. Those 4 disks can be run just like the were internally as the same RAID 10 set up by the same software RAID program. [ Assuming the software RAID program didn't 'hardcode' the physical device (e.g., a disk would only work in drive sled 1 ). Most don't including Disk Utility's. ]

There are also relatively affordable RAID cards that just do RAID 0,1, 10. Some folks on these forums spin these as not being "real" RAID cards. If only want to do RAID 10 they are "real enough". However, their hardware assisted/dedicated RAID layout will be different from the software RAID. It is like changing to a different RAID vendor. You'd have to reformat the whole RAID 10 array and then refill the data.

This is also a little confusing to me. How do I figure out if the workload switches to the video card?

Things get done quicker but the measure CPU load drops then the load switched. Example would be exports decreased in time but the CPU load went down. OpenCL code doesn't "have to" be dispatched to the GPGPU. That is one of the differences with CUDA (what Adobe Premiere has a high acceleration dependance upon ) which can only dispatch to Nvidia hardware only. If the GPU is swapped OpenCL code can be dispatched to the x86 cores.

For FCPX some of the defects for in the context of multiple cards suggests that either the Mac OS X opencl core support or the app's OpenCL code makes an assumption that the app's GPU card is the only place to dispatch too.

Also, I guess I'm a little bit lost about where all the drives would end up. If I end up sticking the the old discs, no problem. But say I opt for a RAID setup with SSDs, where do they go?

It is questionable you need to RAID SSDs. But yes if put four more drives inside the Mac Pro you can move the old ones outside. If you want fast access to the HDDs outside you can get a SATA/SAS/RAID card and an exterenal enclosure. If don't need/want high speed you can use something like one of these "raid in a box" enclosures:

http://eshop.macsales.com/shop/hard-drives/RAID/Desktop/

Takes four drives. (May want to double check that the drives are on the "compatible" list). You can set up external RAID 10 for $280 and the exact same drives, hanging off the FW800 port if don't need speed.

It is bit unclear where these "old" projects go when "cleared out". If they are mostly thrown away then it seems there was some external solution already.

But still, at least for now, I think I'll go with the incrimental upgrades: RAM, GPU, switch from RAID 10 to RAID 0, and then see how that goes.

From a safety perspective, RAID 0 is a downgrade; not an upgrade. Going from '10' to '0' you loose something. That is why the digits change.

You can put the safety back in by frequently mirroring ( Time Machine, back-up ) to an external RAID-10 set-up but that is a substantial increase in the the number of disks involved in the system to largely get back to where you were with the original four.

 

[mikey1004]

Wow! Do you get paid to write this stuff? If not, you certainly should. This is a tremendous help. Let's see if I understood all of it.

RAID 0 group 1 -- the sources
RAID 0 group 2 -- the outputs

So if reading and then streaming back out that is being written to two different sets of spindles. That will speed up throughput more than effectively binding 3 spindles together as a group.

...

Many Apple products tend to want to heap everything into one big pile inside of a single folder. Having to deal with data in multiple locations will actually exposure more features of FCPX than letting it do the automagic "copy and merge into blackhole" default mode.

So by "sources," do you mean like the source clips for the movies? I.e., as FCPX calls them, "Events"? And then "outputs" would be the completed, exported movies, right? Your explanation as to why this would improve things is quite clear. But then where would the Final Cut "Projects" go? With the sources or with the outputs? Also, I assume that "groups" are separate RAID arrays, right? So we essentially would have two separate RAID 0's, each consisting of two disks.

That said RAID 0 carries risks that are best laid out for the user so they understand the trade-off they are making. It will take much longer to clean up the mess after a failure. They will likely loose what they were working on just before the failure. If reworking effort and being "down" for a day or so is OK, then it is a reasonable trade off. If not then should avoid RAID 0.
The fact the original set up is not RAID 0 implies that at some point that safety factor was a concern.

Understood. The reasoning behind choosing RAID 10 in the first place was not so much safety so much as simplicity for the user: I figured it would be easier for my brother to work with only "one" HDD (other than the OS/apps disk), but I didn't trust him to plug in the USB HDD's that he uses for backup and run Time Machine with sufficient regularity. So I felt that redundancy in the system was necessary. However, now that he is more comfortable with the machine, I'm sure he can deal with two disks, and make sure his Time Machine is always plugged in.

I'm not sure that editing 2D animation is really going to stretch FCPX for enough to justify the second 5770 just to "unload" the main 5770 for other apps and their OpenCL processing. There are some subcomponents of FCPX that heavily leverage OpenCL and others that don't. However, since he wants to do everything on a just one monitor dual 5770's won't buy much. [ FCPX doesn't work so well when the program's windows are split over two video cards. However, can make better use of the single card if can move other GUI and/or GPGPU workload off. ]

Just one 5770 is would be a significant improvement to the the single GT 120 he is running FCPX on now. A 5870 would be a margin improvement over that. Not sure really getting "bang for the buck" with that marginal improvement; there is a $200 gap between the cards. So a 80% increase in price for often nowhere near 60-70% increase in performance. (on paper it is a 100% performance increase but that will only show up in corner cases ). Both cards are discontinued in the Windows PC market. If there "extremely tight" project deadlines it might be worth it, but as a hobby project. No. That money could be much better spent on other things (not necessarily hardware) that has a much better return on investment.

Ah, but you see, he doesn't use FCPX for his cartoons. In fact, all of his cartoons are hand-drawn and edited at an animation studio. He almost never uses his Mac Pro for that. FCPX is used for live-action movies that he films in full 1080p (often with two cameras simultaneously). Still, though, it sounds like just the one 5770 will be sufficient.

And that is the only project he is going to work on for the next 1-2 years? That project will end too. And then can do another set of upgrades. "Project end gaps" are not a good justification for even larger "Big Bang" migrations.

You know, that's a great point. One step at a time.

I'm saying you don't need a fancy, very high speed RAID card to put 4 disks in an external enclosure. There are cards which will give you eSATA or SAS/SATA external connections that don't do any "RAID". You can just get an external connection to 4 disks. Those 4 disks can be run just like the were internally as the same RAID 10 set up by the same software RAID program. [ Assuming the software RAID program didn't 'hardcode' the physical device (e.g., a disk would only work in drive sled 1 ). Most don't including Disk Utility's. ]

There are also relatively affordable RAID cards that just do RAID 0,1, 10. Some folks on these forums spin these as not being "real" RAID cards. If only want to do RAID 10 they are "real enough". However, their hardware assisted/dedicated RAID layout will be different from the software RAID. It is like changing to a different RAID vendor. You'd have to reformat the whole RAID 10 array and then refill the data.

...

It is questionable you need to RAID SSDs. But yes if put four more drives inside the Mac Pro you can move the old ones outside. If you want fast access to the HDDs outside you can get a SATA/SAS/RAID card and an exterenal enclosure. If don't need/want high speed you can use something like one of these "raid in a box" enclosures:

http://eshop.macsales.com/shop/hard-drives/RAID/Desktop/

Takes four drives. (May want to double check that the drives are on the "compatible" list). You can set up external RAID 10 for $280 and the exact same drives, hanging off the FW800 port if don't need speed.

I like that "RAID in a box" idea. That way we can stick the old drives in that and use them as a better backup than those USB things we have. The enclosure you linked, is that an example of a not-quite-real RAID card, as you described above? If so, it sounds plenty sufficient for Time Machine use.

It is bit unclear where these "old" projects go when "cleared out". If they are mostly thrown away then it seems there was some external solution already.

Yes, they are literally erased from existence. All that remains are the completed, exported movies--all the Event and Project files are deleted.

From a safety perspective, RAID 0 is a downgrade; not an upgrade. Going from '10' to '0' you loose something. That is why the digits change.

You can put the safety back in by frequently mirroring ( Time Machine, back-up ) to an external RAID-10 set-up but that is a substantial increase in the the number of disks involved in the system to largely get back to where you were with the original four.

Hmm. I understand your point. But, if I understand your explanation, I do think your proposed idea of two RAID 0's will improve the workflow. Actually, now that I think about it, I have no idea how I set up the RAID 10 in the first place, so that could be the root of the reported slowness that my brother is experiencing (I feel like there was a crappy youtube video involved...).

Whatever the case, maybe starting over with some new disks will hold him over for a couple of years, that is, until SSD's really do become a viable solution. I assume Caviar Blacks still the cream of the crop, yeah?

So it looks like the plan is as follows:

1) Upgrade RAM from 12 GB to 32 GB
http://eshop.macsales.com/item/Other%20World%20Computing/85MP3W4M32GK/

2) Upgrade GPU from 3 x GeForce GT 120 to 1 x ATI Radeon HD 5770
http://eshop.macsales.com/item/Apple/6615718/

3) Move 4 x 2 TB HDD's to external enclosure, rebuild RAID 10, use as Time Machine. (Or maybe RAID 5 for additional space for Time Machine history?)
http://eshop.macsales.com/item/Other%20World%20Computing/MEQX2KIT0GB/

4) Purchase 4 x 1TB WD Caviar Blacks. Create 2 x 2 TB RAID 0 arrays, one for sources, one for outputs. Transfer data onto them from Time Machine.

5) See if that's enough. If not, maybe look at CPU upgrade options.


Waddaya think?

 

[wonderspark]

I agree that RAID 0 is a downgrade from RAID 10 generally speaking, but some key points made here raise some questions.

One is that you said you didn't trust him to back stuff up with Time Machine, so you made a RAID 10 for redundancy. The problem with RAID 10 (in my opinion) is that if you make a mistake in deleting something, it's gone from both volumes, and so it's *not* redundant for erroneous deletion. It *is* good for the case of one drive failing, however.

I also agree that two sets of RAID 0 with source files on one set and output files on the other can be better than one set of three, if you're doing all the source and output on that same RAID set. My suggestion was based on the possible need for pure speed, in which case the three or four disks in RAID 0 could be used as the source, and output sent to an additional disk outside the RAID, either the fourth disk, or an external disk. Generally, you don't need that much throughput to write an output file, as opposed to reading multiple layers of source files. This was how I initially used my Mac Pro when editing a movie shot on DVCProHD, and it worked flawlessly on three 1TB disks in RAID 0.

I may have just been lucky, but I don't have a lot of fear associated with RAID 0. I've always done my backups either after a really critical point of work or at the end of an editing session, and always to an external disk or backup apart from working disks, and have not been stung by failure mid-edit. I've only had to replace one disk ever, in fact, and it gave me plenty of warning prior to complete failure. Western Digital RMA'd me a new replacement prior to the complete failure of the disk, and I sent them their intermittently failing disk after swapping in the new one, so it was as smooth as could be.

One thing I still don't understand is why replace four 2TB disks with four 1TB disks? If you're using the extra disks for backup, that's good, but I would advise *against* a RAID 0 (stripe) array for backups, which should be on single disks in my opinion. You could run the two RAID 0 sets for source and output, then use the external disks for backups as individual disks, and get the benefit of being able to grab the external disk set and run in case of fire or flood.

Also, it's better to only use 50% of any disk or RAID set, so that you're accessing the fastest part of the disks (where the rotational speed is greatest on the outside of the disk(s)). So using the 2TB disks for the internal RAID set at 50% capacity gives the greatest throughput, while staying within the size limits of the four new 1TB disks, if you follow.

 

[deconstruct60]

So by "sources," do you mean like the source clips for the movies? I.e., as FCPX calls them, "Events"? And then "outputs" would be the completed, exported movies, right?

Partially. If you are looking for two labels to put on the two Volumes one would be "Outside_FCP" ( Events, but also the Output that will be played outside of the application) and "Inside_FCP" ( stuff in formats that FCP likes more. ).

While better at it now FCP typically wants to transform the original material into ProRes format. So that is one source -> output move. Then at the end of the process there is a transformation from ProRes to output format (since nobody really watches anything in that format).

So which set is the "source" and "output" is a bit fluid at different stages. It is more illuminating to note there are two which can be directed at different volumes. Since you have 4 drive sleds, that can be two sets of two. [ The other comment above is right that the final destination doesn't have to be RAID , but can switch which way things are going to split the load in each phase. ]

There are settings to leave the originals alone in place (do not copy Events), but FCPX will want to generate either ProRES or a proxy in the same Volume as the project. Some more background:

http://www.larryjordan.biz/media-in-fcp-x/

[ In some corner cases FCPX will deem the originals as "good enough" to work with and won't duplicate to edit. May want to switch up what the target disk in that case (if all "good enough" ) or pick a third volume. ]

Your explanation as to why this would improve things is quite clear. But then where would the Final Cut "Projects" go? With the sources or with the outputs? Also, I assume that "groups" are separate RAID arrays, right? So we essentially would have two separate RAID 0's, each consisting of two disks.

The "Projects" go on the opposite one from the raw events. The "Inside FCP" set. Seems a bit of a leap to label them as "arrays" since using software RAID but, yes virtual arrays each with two.

Understood. The reasoning behind choosing RAID 10 in the first place was not so much safety so much as simplicity for the user:

If keeping the simplicity approach labeling them "Outside_FCPX" (or FCPX_Evnents ) and "Inside_FCPX" ( or FCPX_Projects ) probably will work better than "source" and "output".

I'm sure he can deal with two disks, and make sure his Time Machine is always plugged in.

It will be 4 since the OS/Apps is on a different drive, but yes.

FCPX is used for live-action movies that he films in full 1080p (often with two cameras simultaneously). Still, though, it sounds like just the one 5770 will be sufficient.

Ah. Well should be in good shape as long as can dual stream the output from the cameras off the one "outside" RAID group. For mainstream 1080p cameras that should work.

I like that "RAID in a box" idea. That way we can stick the old drives in that and use them as a better backup than those USB things we have. The enclosure you linked, is that an example of a not-quite-real RAID card, as you described above?

Yeah only the card is effectively in the box. I would not use that box for RAID-5 though, but in RAID-10 mode it would be fine.

[ "real RAID" would have option for battery back-up, RAID 6 options , better control over the array, etc. ]

If so, it sounds plenty sufficient for Time Machine use.

Should work.

Once caveat would be that if do Time Machine back-ups of the working drives it would be good to go "clean up" the back-up images after finish projects and don't need those back-ups anymore. Time Machine has the "bad" habit of expanding to completely filling whatever volume it resides on. ( it just keeps growing ... sometime past where you might want it to. Once it hits the barrier it starts to erase stuff automagically. What it starts to erase may not be in the priority order you would like. )


Time Machine for the OS/Apps/Users drive probably will work better.

Something closer to rsync ( Crashplan, SuperDuper , CCC, ? ) might work better if just want to mirror the working spaces to the external "mirror/clone". That isn't an archival back-up but at least gets you one "point in time" image to fall back to.

Perhaps

- 200GB partition for OS disk as a Time machine target
- Rest for a clone of the projects RAID set (assuming the original sources are on other disks somewhere. )

Yes, they are literally erased from existence. All that remains are the completed, exported movies--all the Event and Project files are deleted.

so it would work to initially target the "outside" set and then copy it off to a more permanent location when sure the project is done.

Actually, now that I think about it, I have no idea how I set up the RAID 10 in the first place, so that could be the root of the reported slowness that my brother is experiencing (I feel like there was a crappy youtube video involved...).

It isn't too hard if just follow some simple directions and use Disk Utility. You just tell Disk Utility you want to use those four disks for RAID and then tag them as a RAID 10 set (from a drop down menu item). It does the rest.
For example:

http://macperformanceguide.com/Storage-HowToSetupRAID.html

There is a software package called SoftRAID ( http://www.softraid.com/ ) which has more higher end RAID admin features ( predicative disk failure, volume validation , Generate bug report , etc. ) , but so far doesn't do RAID 10

Whatever the case, maybe starting over with some new disks will hold him over for a couple of years, that is, until SSD's really do become a viable solution.

SSDs are viable now if have extra $1500-2000 lying around. LOL.

I assume Caviar Blacks still the cream of the crop, yeah?

There are some new Velociraptors (VR) :

http://www.storagereview.com/western_digital_velociraptor_1tb_review

[not sure why it hiccups on the HTPC benchmark. ]

http://www.anandtech.com/show/5729/western-digital-velociraptor-1tb-wd1000dhtz-review/


but that is a slippery slope into a price region close to some of the SSD clearance sales going on now. However, some WD RE4 are quite suitable since limited to the Mac Pro SATA II (3Gbs) controller. There is firmware tweaks on those to deal better in a RAID context (e.g., vibrations and errors ) so not exactly the same as the Blacks.

At some point there must have been a price premium, but at this point, the RE4's appear to be priced about the same a Blacks. Since it is actually a RAID context going to use them in, that makes for a good match.

It is a bit uneven but an alternative:

2 x 500GB new VR ( for "outside" )
2 x 1TB RE4 (for "inside" )

would be 3TB. Which is good because it is smaller than the 4TB the old RAID-10 presents as. It also is avoiding the 1TB VR due to pricing (pretty close to a SSD of half the size). Not sure if the original footage would be over 1TB, but getting two streams off the Velociraptor at the same time may work better.

A more expensive option would be

2 x 500GB RE4 ( for "outside")
2 x 1TB VR ( for "inside" )
Depends upon if expect the streaming bottleneck in the transform process or when pulling two ProRes streams.

Under ideal conditions the new VR can sequentially move around 200 MB/s
Uncompressed 10-bit 1080p video is around 125 MB/s.

http://www.matrox.com/video/en/support/mac/mxo2/design/storage/

Probably not going to be streaming that, but if pull two streams from one drive concurrently you won't see all of that either. So a bit under half is a conservative guess which is pretty close. Throw it into RAID 0 and should be pretty good shape.

So it looks like the plan is as follows:

1) Upgrade RAM from 12 GB to 32 GB
http://eshop.macsales.com/item/Other%20World%20Computing/85MP3W4M32GK/

6 x 8GB 24GB would probably work too.

Could pick up 2 x 8GB later if need be.

2) Upgrade GPU from 3 x GeForce GT 120 to 1 x ATI Radeon HD 5770
http://eshop.macsales.com/item/Apple/6615718/

fine.

3) Move 4 x 2 TB HDD's to external enclosure, rebuild RAID 10, use as Time Machine. (Or maybe RAID 5 for additional space for Time Machine history?)
http://eshop.macsales.com/item/Other%20World%20Computing/MEQX2KIT0GB/

Can consider picking up one of their PCI-e eSATA cards if want to have faster access to the storage.

4) Purchase 4 x 1TB WD Caviar Blacks. Create 2 x 2 TB RAID 0 arrays, one for sources, one for outputs. Transfer data onto them from Time Machine.

Have a variant above.

5) See if that's enough. If not, maybe look at CPU upgrade options.

One way of doing of improving if not the CPU would be a PCI-e SSD. It would not have to be huge. For example the 128GB OWC Accelsior, but there are likely some better, cheaper options in the future.

But should be OK with the internal controller with just 5 larger storage devices on it.

 

[deconstruct60]

One thing I still don't understand is why replace four 2TB disks with four 1TB disks?

There is a part in the original post where brother is willing to trade off size for space. The current RAID 10 set presents as 4TB. He thinks he can get by with 2TB as a working drive and will to trade that 2TB of usable space for more speed.

Two sets of 2TB would actually end up with more usable space (presenting as two 4TB Volumes). If you try to mirror/back-up both of those you have more space than the back-up drives ( composed of ). Each individual set is also bigger than any single drive in the back-up set (if break the up from RAID 10).

If have two internal sets that present as 2TB then conceptually could have two external pairs that presents as 2TB. A Time machine drive that is mirrored is probably better. The fact they are same size not so much if the internals are filled to capacity. ( I only really use about 1TB of those internal drives then in much better shape. That would be good since the file system on those drives would likely stay less fragmented also . )

If you're using the extra disks for backup, that's good, but I would advise *against* a RAID 0 (stripe) array for backups, which should be on single disks in my opinion. You could run the two RAID 0 sets for source and output, then use the external disks for backups as individual disks, and get the benefit of being able to grab the external disk set and run in case of fire or flood.

Striping isn't so much the root cause of the problem as no duplication ( mirror/parity) is. The problem with "only use individual" disk mantra is that back-up size is limited to whatever max external capacity currently is. If want to keep archived alternative files that only makes the limitation more pronounced.

So using the 2TB disks for the internal RAID set at 50% capacity gives the greatest throughput, while staying within the size limits of the four new 1TB disks, if you follow.

"Short stroking" 2TB drives may not be as good as just getting faster 1TB drives. It can be less expensive if use a different class ( Black/RE vs Velociraptor). But it is used in the same class too. The new 500GB VR is reported to be just as short stroked from the factory 1TB drive ( so should have a bit better random I/O performance.).

 

[wonderspark]

Those VR disks *are* pretty fast.

As far as using single disks for backups, I do like the Hitachi 4TB 5K4000. It's been fantastic for me so far, and they seem to be going for $350 lately.

 

[deconstruct60]

As far as using single disks for backups, I do like the Hitachi 4TB 5K4000. It's been fantastic for me so far, and they seem to be going for $350 lately.

The problem is that single disks mean single point of failure. So it is a dual edge sword. Multiple cheaper disks can be "merged" to present larger storage, but have more complexity ( custom disk data formats. etc. ). On the other hand, more expensive single disks enable more mainstream recovery mechanism/tools, but have just shifted the single point of failure.

If the "back-up" process is non-archival (meaning primarily just have a copy of the latest so can restore to what you have ) then the primary working source can be the "back-up of the back-up". The common solution mode is that if Time Machine disk fails then you put in a new clean disk and start another Time Machine process.

The problem there is that the "back in time" data that Time Machine held is now largely gone. (there may be some few fragments left on the original source volume, but it is mostly only stored in the Time Machine image on the target volume. ).

Pragmatically, that is probably OK for most folks. Lots of folks turn on Time Machine and never dive into the archives, let alone recover something from it. Similarly, the "oops I deleted a file two hours ago" recovery is already done if there is a archive failure a month later.

The problem will run into once the single disks more common get into the 4TB range is that several of the recovery tools folks have used in the past start to have problems. For example, Spinrite (leveraging DOS to do low level peek/poking) and Drive Genius (map the file system metadata to 32-bit memory space to debug) have scale problems. DOS can't deal with that capacity size and sometimes 4TB of data can have multiple GB sized metadata (especially Time Machine is its vast array of hard links to "duplicate" files. )

 

[wonderspark]

I don't like or use Time Machine. I just copy files I need backed up onto external disks. If it's really critical, I back it up twice (three copies total) on yet another disk. My intent for the backups are to restore something lost on my working drives, whether by failure or accidental deletion. That's why a single backup disk works for me. Fortunately, I've not needed them thus far.

 

[mobilitech]

Wow, I learned a lot in this thread! But I think I will have to make my own thread for storage issues…

 

[xaviercounord]

More info on your set-up ?

Hello
Could you tell me more about your set-up. I understand that your RAID is external why did you get this specific card, do you use internal SATA disk ? Is it complex to replace the CPU ? who are the supplier ?
I have a quad core 2009 mac and just order 32 GB of memory to replace the initial 3GB. I believe I should see a difference (photo processing, large library of photography) however I would like to increase disk access and have ruminated about internal solution with no a priori expertise
Looking forward to hearing from you
jc Ottawa



I have a RAID that I dumped some money into, but it's pretty fantastic. It's an 8-bay external box (by Sans Digital) filled with 8x Western Digital 2TB RE-4 disks in RAID 6 via an Areca 1880ix-12 card. It gives me 12TB of storage with the safety of allowing any two of the eight disks to fail without losing any data. It provides read speeds of 714MB/sec and write speeds of 816MB/sec based on my own tests, and I can even edit native, unrendered DSLR footage while rebuilding the RAID during a disk failure. (Another test I did by pulling a drive while editing.) You might be able to build something with similar speeds using SSDs, but the storage capacity will not come close, however he doesn't seem to need capacity so it's an idea. I'm personally not ready to trust an SSD RAID just yet, and my livelihood depends on my system.

I think the 2.26GHz CPUs are slow, and if you're brave enough to swap in new CPUs, it can be done with some specific sized washers to compensate for the difference in height between lidded and un-lidded CPUs. It would make a huge difference.[/QUOTE]

 

[wonderspark]

Hello
Could you tell me more about your set-up. I understand that your RAID is external why did you get this specific card, do you use internal SATA disk ? Is it complex to replace the CPU ? who are the supplier ?
I have a quad core 2009 mac and just order 32 GB of memory to replace the initial 3GB. I believe I should see a difference (photo processing, large library of photography) however I would like to increase disk access and have ruminated about internal solution with no a priori expertise
Looking forward to hearing from you
jc Ottawa

I have a RAID that I dumped some money into, but it's pretty fantastic. It's an 8-bay external box (by Sans Digital) filled with 8x Western Digital 2TB RE-4 disks in RAID 6 via an Areca 1880ix-12 card. It gives me 12TB of storage with the safety of allowing any two of the eight disks to fail without losing any data. It provides read speeds of 714MB/sec and write speeds of 816MB/sec based on my own tests, and I can even edit native, unrendered DSLR footage while rebuilding the RAID during a disk failure. (Another test I did by pulling a drive while editing.) You might be able to build something with similar speeds using SSDs, but the storage capacity will not come close, however he doesn't seem to need capacity so it's an idea. I'm personally not ready to trust an SSD RAID just yet, and my livelihood depends on my system.

I think the 2.26GHz CPUs are slow, and if you're brave enough to swap in new CPUs, it can be done with some specific sized washers to compensate for the difference in height between lidded and un-lidded CPUs. It would make a huge difference.

I got the 1880ix because it was a pretty good price, and I wanted to have options for expansion, both internally and externally. I currently run my RAID box externally through internal ports on the card via 8087-->8088 cables, which isn't the cleanest looking solution, but obviously works fantastically. I still have one internal and one external port to work with, which can give me a full 8 more disks to RAID for 16 internal SATA disks total, if needed. So far, that need has not come, and it's possible that the 1880x would have been all I ever needed, but the price for that card was so close to what I paid for mine, that it seemed silly not to buy the stronger one.

Replacing the CPU was not complex at all. I was pleasantly surprised with how simple it was. Key is to prepare all your information, get the proper tools and supplies, and follow directions from those that have already done it before you. Everything just clicks, pops, slides, lifts, drops, and snaps back together without any drama. I got the CPU from Provantage, and if you google w3680 provantage, you can't miss it, currently $596.54.

32GB of RAM in four matched sticks of 8GB each will definitely improve your work. Set the original RAM aside for backup or troubleshooting.

If you need links to photo instructions, I'll dig them up, but it's very easy to find them on this forum with search.