MacPro RAID Card

[Bob]

I have just received and installed the MacPro RAID card and here are my first impressions:
Installation:
Pretty straightforward, but required removal of memory, video card, all disc drives and some sheet metal. I was surprised to find that the disassembly differed based on a four core or eight core machine. Apple seems to be designing each machine as a one-off.
The reason for most of this disassembly was to get to the connector on the motherboard that attaches a cable going to the drive connectors. Besides disconnecting this cable from the motherboard and attaching it to the RAID card, the process involved re-routing this cable to provide the reach necessary to get to the connector on the RAID card. This was a bit of a PITA and seemed impossible until I removed and re-installed the connector for drive 1 so as to route the cable to the right of the connector. Note that this move is not in the manual.
After the hardware assembly, I installed four Hitachi 500 Gig drives which are currently $110.00 each from newegg.
My RAID configuration is a single three disc raid 5 RAID set with the fourth drive configured as a hot spare.
I formatted the raid set as a single volume. The RAID card took about five hours to construct the raid set and another couple of hours to fully change the on-card backup battery.
After restoring from one or my previously made backups (using the disc utility). I was up and running.

Performance:
This is pretty impressive, especially when the file fits in the RAID's on-board buffers. Photoshop opens in three seconds. I am using this configuration mainly for reliability, and it is pretty nifty that reliability and speed arrive together. You could go faster by striping all of the drives, but nearly double is fine for me given I am now protected from disc failure. Smallish files read very rapidly, due to some predictive buffering I think. Largish files (of the one gig size or so) read at about double speed.

Is it worth the money?
$1000.00 is a lot, but I think of it as data insurance.

Other considerations:
I am continuing to use Leopard's time machine as well as an occasional external backup.
All of my really important stuff gets backed up off site via usb drive or the internet.

-bob

 

[Daniel]

bob,

i installed it too. i think this was the most difficult apple installation i've done for myself. i actually read the manual VERY carefully. quite unusual for me.

i'm most concern with hardware failure, so i've set mine up as two separate mirrored drives. all four bays are occupied.

anyway, it's expensive, but i think it's worth the investment in redundancy. but did you know that Mac OS prior to 10.5 will not recognize the RAID card?

—dan

 

[Jack]

What size disc are you guys dedicating for Time Machine?

 

[Bob]

I use a 1TB LaCie striped unit attached via FW800. I also exclude Lightroom catalogs and VMWare virtual machines. To prevent long stalls, I also set iDisk for manual sync.
-bob

 

[Daniel]

What size disc are you guys dedicating for Time Machine?

i've had hardware failure before where i lost over 12 months worth of hard work in only several seconds, but it seemed like eternity, so i'm overly cautious with hardware redundancy.

anyway, all four bays are taken up by 750 gig drives. one pair is set up as a raid one, and the other pair the same. so i have two raid one drives. the second one is used as an emergency backup boot drive. i also have a 500 gig external firewire drive (Lacie), and an external raid one drive (G Speed eS by g-technology). both are used alternately as time machine drives.

regarding time machine. you should have at least a 250 gig drive. the size of your time machine drive depends on the average size of the total contents of your system and home folders.

my time machine drives are between 500 and 750 gigs. the larger your time machine drive, the further back in history your time machine can go back to recover accidentally deleted or lost files. however, when your time machine starts running low on disk space, time machine will remove old files in history to make room for backing up recent files.

i hope this made sense.

— dan

 

[Jack]

With your twin 2xRAID1 set-up, what drive are you using for CS scratch, or do you not bother?

 

[Jack]

I use a 1TB LaCie striped unit attached via FW800. I also exclude Lightroom catalogs and VMWare virtual machines. To prevent long stalls, I also set iDisk for manual sync.
-bob

Hi Bob:

So I assume there is a setting in TM that allows you to exclude certain folders or files? I ordered a year of .Mac with my new machine, so I'll probably be picking your brain on that in Moab

 

[Jack]

I use a 1TB LaCie striped unit attached via FW800. I also exclude Lightroom catalogs and VMWare virtual machines. To prevent long stalls, I also set iDisk for manual sync.
-bob

Hi Bob:

So I assume there is a setting in TM that allows you to exclude certain folders or files?

Re iDisk, I ordered a year of .Mac with my new machine, so I'll probably be picking your brain on that in Moab

Cheers,

 

[Bob]

I try to have enough ram so that the photoshop scratch disk is rarely/never used. 6GB seems to be enough.
My raid configuration id three drives configured into a raid 5 with the fourth as a hot spare. I view raid as a way to increase reliability (the R in Raid) with a modest increase in speed.
The speed increase occurs because effectively the three disks are two-way stripped, with the third being the checksum (actually true for any two of the three you pick.
It is more valuable to me to have the hot spare for rebuilds should one of the three fail, then it is to get a little more speed.
I use iDisk for several things, but not for holding my raw photo library.
I plan to arrive Sunday Feb 3 in the late afternoon and will be carrying Leica or Nikon depending on New Jersey getting at least one body back to me.
-bob

 

[DougDolde]

Actually the R in Raid is Redundant.

 

[Bob]

Not when we coined it

 

[Jack]

I try to have enough ram so that the photoshop scratch disk is rarely/never used. 6GB seems to be enough.
My raid configuration id three drives configured into a raid 5 with the fourth as a hot spare. I view raid as a way to increase reliability (the R in Raid) with a modest increase in speed.
The speed increase occurs because effectively the three disks are two-way stripped, with the third being the checksum (actually true for any two of the three you pick.
It is more valuable to me to have the hot spare for rebuilds should one of the three fail, then it is to get a little more speed.
I use iDisk for several things, but not for holding my raw photo library.
I plan to arrive Sunday Feb 3 in the late afternoon and will be carrying Leica or Nikon depending on New Jersey getting at least one body back to me.
-bob

Super --- we'll see you when you get in!

On the RAID, I totally get the logic of RAID 5 for redundancy and speed for moving and storage of critical large files like images, and would love to see a FW800 or GigLAN version of a DROBO for my images. But not sure the RAID5 benefits warrant taking up three internal slots for the OS drive, especially with programs like Time Machine for back-up and especially now that drives seem to be more reliable in general...

 

[Bob]

I used to care, and my servers used to be configured with a mirrored boot drive pagefiles on a stripe set, and critical data on the raid 5, but now with single drives holding a TB each and a four slot maximum on the raid card, I just configure one large raid 5 over three slots and keep a slot filled with a hot spare.
Now if the MacPro had eight to ten drive bays I might come up with a more complicated preferred configuration.
It makes no sense to me to divide a physical drive into more than one role. I have seen some folks advocate partitioning the data storage raid drives into a data storage segment and a raid 5 segment. Drives tend to fail at the sector/record level or at the head level or at the total drive level, with the probability of total driver failure coming in second to single sector/record level. When the drive fails, then all participating volumes are affected so no improvement in reliability is achieved. In fact system reliability is worse since a single failure takes out more than one system component.
-bob

 

[Lars]

Bob,

I learnt the hard way a week ago that my controller wouldn't let me replace a failed drive on my RAID5 array. I lost the data on the array. I have a backup offsite, but that's the other end of Europe since I'm offsite over the winter. I went cheap, didn't buy that extra 1TB drive for onsite backup. :cussing:

Three lessons learned:
1. Keep an onsite as well as offsite backup. Should be an obvious thing but this needs to be reiterated.
2. Test BEFORE you start trusting the array that it will actually allow for replacement of a failed drive. Start a rebuild and pull one drive in the middle of the process. It should accept a new drive and should allow you to access data on the array.
3. The controller card can fail as well. RAID arrays use proprietary disk formats, another card cannot be expected to read the data. Always buy controller cards in pairs, and test both before deploying.

My own opinion is that RAID5 is for NAS-type storage whereas 0+1 is better for a workstation.

 

[Bob]

Have it your way...
I always check that the setup does what it is supposed to do. Better now then at a failed recovery attempt.
I hate to throw away recovery options since they cost so little;I have also had times when my external HD could not be read.
So belt suspenders and an insurance policy are all in my daily routine.
-bob

 

[Daniel]

With your twin 2xRAID1 set-up, what drive are you using for CS scratch, or do you not bother?

i use the 2nd raid-1 drive as my CS scratch disk. but i don't notice much difference because my photoshop files do not get very large. my photoshop files are usually between 50 and 100 mb.

i would assume that whether it makes a speed difference between the boot drive and the separate drive as your cs scratch disk will depend on how large your photoshop files become because of all the different layers.

hope this made sense.

—dan

 

[Jack]

Yeah, my files often get really big while I'm working on them --- 16-bit and several layers, and it's not uncommon to get files up around 600 - 800 megs.

 

[Lars]

With CS2, PS scratch files for 16-bit editing are many times larger than with 8-bit files. While retouching large scans from 8x10 I often see scratch file sizes in the range of 15-25 GB - 10-20 times the image size. In 8-bit mode the scratch file is much smaller. Has this been fixed in CS3?

 

[Jack]

Not really, but then that isn't a true scratch size, just the reserved amount. The problem is if you have several images open at the same time, the reserved scratch for all of them is additive, and so you need a fairly large scratch space to accommodate it. The other (STUPID!) issue with CS3 scratch is the files do NOT purge on the fly, but only after shutting the program down and re-starting(!) So basically, the sector of scratch reserved for an image you closed a half hour ago is NOT available on the drive for any further scratching in that CS3 session.

Here is where excess RAM is beneficial, because it does purge on the fly. Bottom line, is Adobe's scratch dialogs are archaic and need a total re-vamp to get in line with current technology.

/rant

 

[Lars]

Hmm, my observation is that PS2 actually does write tens of GB to the disk when editing 16-bit (huge) images. I notice this when loading a 16-bit 2000 dpi 8x10 drum scan and converting to a working color space - we're talking several minutes of writing to scratch file, on a 4GB system with 2.2GB available for PS. In 8-bit mode processing time is almost nothing and the scratch file is tiny. Major difference between 8 and 16 bit mode, much more than a factor of two.

Same thing when converting a layered masked PS file from 8-bit to 16-bit. I think Adobe never got around to optimizing 16-bit mode in CS2, thus my question about CS3.