Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Please critique: My plan for configuring 4 hard drives in Mac Pro

J

jasone6

macrumors member
Original poster
Jan 9, 2008
94
0
I've had my Mac Pro (8-core, 2.8GHz) for almost a year. I just purchased two more hard drives to fill the remaining 2 bays -- for a total of 4 one-TB drives. My data is backed up, and now I'm ready to start completely fresh. I've done some research -- much of it at diglloyd and I'd like some feedback on my plan.

I put together a graphic to show what I've got in mind. A couple things to note:

- I prefer to have an internal Time Machine backup
- I know 2-drive striped RAIDs double chance of failure
- I know Time Machine should be 1.5x (or more) the size of the data volume, but I don't plan to use all the data volume, so it should be fine.
- One drive is Hitachi (from Apple), and remaining 3 are Samsung F1

Aside from general reactions, I'd love to know if there are any special considerations in setting up my striped RAIDs. I've never set up RAID configurations before. Partition Table considerations?

Thanks!

-Jason

MP HD config (low).jpg
 
Just my option...
Having two sets of striped drives doesn't make for good redundancy, if your gonna use the additional drives for backup it makes more sense to keep them as single volumes.

Seriously, using stripe drives is a accident waiting to happen.
The only benefit from having striping is speed!!
 
I don't think the plan is good at all. Using RAID 0 these days makes no sense. Especially for a Time Machine volume. You are just asking for data loss.

I would instead opt to use RAID 1+0 (mirrored then striped) using the four 1TB drives. Then I would get an external 1TB or 1.5TB drive for Time Machine.

Ideally, you will have a spare 1TB around in case a drive in the RAID 1+0 array fails.

Heck, I think you would better off going RAID 1+0 with no time machine than with what you originally planned to do.

S-
 
While the performance isn't the greatest, I have been using RAID5 in my Mac Pro. You could use it with 3 or 4 drives. I guess since you want an internal drive TM, you could use 3. You would basically get (N-1)*Smin storage space .. so in your case 2TB of useable space. So if you limited what you had TM backup, you could use the 4th disk as the backup.

It seems that RAID6 is becoming more recommended over RAID5 in a few articles I have read.
 
gibbz said:
<snip>

It seems that RAID6 is becoming more recommended over RAID5 in a few articles I have read.
Click to expand...

Which articles? There are far superior redundancy setups like 1e or 10a1 etc rathern than RAID6 which has been around as long and RAID5 and yet has not been adopted en masse.
 
UltraNEO* said:
Just my option...
Having two sets of striped drives doesn't make for good redundancy, if your gonna use the additional drives for backup it makes more sense to keep them as single volumes.

Seriously, using stripe drives is a accident waiting to happen.
The only benefit from having striping is speed!!
Click to expand...

I understand that having a 2-disc striped RAID for the "data" volume makes for twice the likelihood of losing data, and I understand that having a similar 2-disc striped RAID for Time Machine means the same.

That said -- for me to lose all my data, 2 separate drives would have to fail at (or near) the same time, right?
- I also have 2 external 500 GB hard drives (for a total of 1-terabyte) backing up my most important data via a non-Time Machine backup program.
- I also use an online backup service for my super-most important data
- I also burn photos, music, etc. to DVD and store in a fire/water-proof safe.

So -- assume that Time Machine is important to me -- but not the end of the world if it fails. Does it make any more sense now?

I've read about other RAID configurations, but its confusing to me. This seems simple enough, moderately reliable, and very fast. Am I off base?

Thanks for the responses...

-J
 
jasone6 said:
I understand that having a 2-disc striped RAID for the "data" volume makes for twice the likelihood of losing data, and I understand that having a similar 2-disc striped RAID for Time Machine means the same.

That said -- for me to lose all my data, 2 separate drives would have to fail at (or near) the same time, right?
- I also have 2 external 500 GB hard drives (for a total of 1-terabyte) backing up my most important data via a non-Time Machine backup program.
- I also use an online backup service for my super-most important data
- I also burn photos, music, etc. to DVD and store in a fire/water-proof safe.

So -- assume that Time Machine is important to me -- but not the end of the world if it fails. Does it make any more sense now?

I've read about other RAID configurations, but its confusing to me. This seems simple enough, moderately reliable, and very fast. Am I off base?

Thanks for the responses...

-J
Click to expand...


You're not really that off base. You can run striped volumes and you will have the same risk of hardware failure as if you ran a single hard drive. The unfortunate compounding result with a striped set is that if one disk goes you lose all data on all disks due to the way the data is written. In some respects, you increase your redundancy by running one drive instead of two in a stripe set.

What you can consider is an external drive with your "absolutely, cannot lose this" data on it. Then go crazy with different setups internally. You'll allow yourself more freedom to play with different setups.
 
Trip.Tucker said:
...The unfortunate compounding result with a striped set is that if one disk goes you lose all data on all disks due to the way the data is written. In some respects, you increase your redundancy by running one drive instead of two in a stripe set.

What you can consider is an external drive with your "absolutely, cannot lose this" data on it. Then go crazy with different setups internally. You'll allow yourself more freedom to play with different setups.
Click to expand...

Thanks Trip. I understand that if one drive fails, I lose all the data within it's counterpart in the striped RAID. Having said that, if I have a second striped RAID acting as Time Machine, for example, then even if I lose a single drive within my main array, I'm still covered (via TM), right? Likewise, if one of the disks in the Time Machine array fails, I lose all of Time Machine -- but I'd still have my main data array, right?

As I understand it, I'd have to lose a drive in EACH of the 2 striped RAID arrays (1 in the main data array, and 1 in the Time Machine array), to lose all my data (notwithstanding what I've backed up to my externals).

I'm having trouble understanding how this would be a less desirable setup than a RAID 1+0. In that setup, I'd also have to have 2 drives fail to lose my data. And, like in the setup I've proposed, there would be a chance that I could get lucky if the 2nd failure came from the right place. (In my scenario, that would be if the 2nd failure was part of the same striped RAID; in the RAID 1+0 scenario, that would be if the 2nd failure was part of the other array. Right?
 
sidewinder said:
I don't think the plan is good at all. Using RAID 0 these days makes no sense. Especially for a Time Machine volume. You are just asking for data loss.

I would instead opt to use RAID 1+0 (mirrored then striped) using the four 1TB drives. Then I would get an external 1TB or 1.5TB drive for Time Machine.

Ideally, you will have a spare 1TB around in case a drive in the RAID 1+0 array fails.

Heck, I think you would better off going RAID 1+0 with no time machine than with what you originally planned to do.

S-
Click to expand...
Thanks sidewinder. I'm interested in your input, but I'm having a hard time understanding why a 1+0 with no Time Machine is better than my proposed configuration. It seems like it has exactly the same probability of total data loss (not to mention the multiple versions of backed-up files that Time Machine keeps).

Thanks again... would love to understand where you're coming from.

-J
 
jasone6 said:
Thanks sidewinder. I'm interested in your input, but I'm having a hard time understanding why a 1+0 with no Time Machine is better than my proposed configuration. It seems like it has exactly the same probability of total data loss (not to mention the multiple versions of backed-up files that Time Machine keeps).

Thanks again... would love to understand where you're coming from.

-J
Click to expand...

With your proposed solution if a drive goes on the first array you will lose your data. So would need to restore it from Time Machine and backups. If one goes on the second array you will need to reinstall your software and OS, and you will lose all your time machine backups.

If you had everything on a RAID 10 setup, then if any single drive went you wouldn't lose anything. You could just put a new drive in and rebuild. If a drive also goes on the other RAID 1 (RAID 10 is two RAID 1s striped) then you would be in the same situation as above, you would just replace both drives. You would need to have both drives from one RAID 1 go to lose data. Therefore it has better fault tolerance.

You would want to use an external drive for time machine (and possibly back that up nightly/weekly to another of the same size for even more redundancy). You of course only get 2TB of usable space on internal drives this way.
 
Why would you have two sets of striped drives? I can understand having it for your main drives, but it makes no sense for Time Machine. It'd be a lot smarter if you had it set up like this:

Drive1 and Drive2= striped
Drive3= time machine of your striped drives
Drive4=mirror of drive3 (or a mirror of drive1 and drive2)
 
Trip.Tucker said:
You can run striped volumes and you will have the same risk of hardware failure as if you ran a single hard drive.
Click to expand...
Not true. A two drive RAID 0 array is twice as likely to fail as a single drive. A three drive RAID 0 array is three times as likely to fail as a single drive. And so on and so on.

S-
 
TuffLuffJimmy said:
Why would you have two sets of striped drives? I can understand having it for your main drives, but it makes no sense for Time Machine. It'd be a lot smarter if you had it set up like this:

Drive1 and Drive2= striped
Drive3= time machine of your striped drives
Drive4=mirror of drive3 (or a mirror of drive1 and drive2)
Click to expand...

Your suggestion would be great, except that Drive1 and Drive2 striped = 2TB, so unless I stripe the 2nd set of drives, I'm looking at trying to backup a 2TB volume with a single TB drive/volume.
 
jasone6,

What you want to build is something that can survive a single drive failure with no down time. RAID 1+0, also called RAID 10, provides that.

In fact, with RAID 10, you could lose two drives, one in each mirror, and everything would still be running. If that happened with your proposed setup, all data would be lost.

You still want to have a Time Machine drive and a spare drive for the RAID 10 array.

S-
 
jasone6 said:
- One drive is Hitachi (from Apple), and remaining 3 are Samsung F1
Click to expand...

This is by far the biggest problem; especially if going with RAID-5.
What it boils down to is that you want the disks to differ as much as possible, given the same size.

The most resilient RAID setup is two or three different brands with disks all of the same size.
If you need more disks than you have brands, try to get disks manufactured from different factories, from different lots, with differing production weeks.

The biggest problem with RAID-5 is that parity (which is used to calculate the real data if it goes missing) is spread out over all disks.
This means that when (no, not if, sadly!) you need to repair the RAID, you're already at a very degraded state, making the remaining disks be read from constantly for *hours*.
This, in turn, makes it much more likely that one of the remaining drives will simply toss in the towel.
I'm glad to say I've used RAID-5 for the last time.
That was with 24 300GB SAS disks (or, well, 9x300 R5, 12x300 R10 and three global hot spares) . The smaller disks and the more solid hardware helped when there was a problem.

If you really want to do something non-wasteful, wait for Snow Kitty and use RAID-Z.
It's less computationally intensive and smarter in every single way, given it's tied to ZFS.

Personally, my next project is likely 6-8 drive AoE storage with ZFS on RAID-10 or ZFS on RAID-Z.
It doesn't get much better than that unless you're using a Centurion to pay for storage ;)
 
rylin said:
This is by far the biggest problem; especially if going with RAID-5.
What it boils down to is that you want the disks to differ as much as possible, given the same size.

The most resilient RAID setup is two or three different brands with disks all of the same size.
If you need more disks than you have brands, try to get disks manufactured from different factories, from different lots, with differing production weeks.
Click to expand...
While I appreciated the idea, it makes very little practical sense. Look at RAID arrays high density storage devices in data centers across the country and you will find that at least 90% of them are built using the same exact drive model. The majority of those are built using Seagate "NS" drives.

I guess it is worth while to try and get drives built at different times. But the reality is that is rarely an issue.

The most resilient RAID setup is built using the best drives available. Most vendors agree that those drives would be Seagate "NS" drives. My personal experience managing a data center for 10 years would back that up. Yes, Seagate had a problem recently with some of their drives. But that is the first time I can recall such a problem and it was resolved quickly enough.

S-
 
sidewinder said:
jasone6,

What you want to build is something that can survive a single drive failure with no down time. RAID 1+0, also called RAID 10, provides that.

In fact, with RAID 10, you could lose two drives, one in each mirror, and everything would still be running. If that happened with your proposed setup, all data would be lost.

You still want to have a Time Machine drive and a spare drive for the RAID 10 array.

S-
Click to expand...

Actually, I don't really think those are my needs. "No down time," would be nice, but isn't my main concern. And while you're right in saying that I could lose 2 drives -- one in each mirror, and still be fine -- I could also lose 2 drives -- both in one mirror, and be fine with my current proposal, right?

I appreciate all the input everyone's been giving. I don't want to buy any more drives or a RAID card. As I see it, my choice is between:

- my proposal: requires 2 drive failures for complete data loss -- possibly 3 (if I get lucky)
- RAID 1+0: requires 2 drive failures for complete data loss -- possibly 3 (if I get lucky). Doesn't provide benefits of Time Machine (i.e. multiple versions of saved files)

I hope I'm not coming across as being argumentative. You guys clearly know a lot on this topic, and I don't. I'm just trying to arrive at an understanding and have yet to see how my proposal is statistically more prone to failure. Any further input is greatly appreciated.

-J
 
2 x 1 TB drives in striped Raid are ca. 1TB in total, aren't they ?

Imho there is nothing wrong with Raid0, in combination with partitioning you get super fast drives for little money, and it's easy to set up.
For batch processing and scratch, it's hard to beat.

The scratch partition I'd keep bigger, depending on your workflow and file sizes you might run out of space there ; data storage using Time Machine might not be convenient, as TM is only doing incremental backups.

My own internal data drives only contain stuff I'm working on, everything else is manually backed up (twice) on external drives when I'm done with it.

Not as save and convenient as a mirrored array, but again, it's simple and affordable.

A fast system/apps drive does not really have many benefits, so a partitioned single drive might be quite fast enough for you.
 
jasone6 said:
I hope I'm not coming across as being argumentative. You guys clearly know a lot on this topic, and I don't. I'm just trying to arrive at an understanding and have yet to see how my proposal is statistically more prone to failure. Any further input is greatly appreciated.

-J
Click to expand...

The big difference between your method and a 4 drive RAID 10 setup is that yours will be backed up and a RAID 10 would provide mirroring.

A drive failure on your "data" stripe would mean all changes would be lost, any new files created etc. since the last time machine backup. With mirroring there is an exact copy at all times.
 
jasone6 said:
Your suggestion would be great, except that Drive1 and Drive2 striped = 2TB, so unless I stripe the 2nd set of drives, I'm looking at trying to backup a 2TB volume with a single TB drive/volume.
Click to expand...

Partition your drives first, then stripe the partitions.

1TB system volume = 2x500GB partitions. Clone this onto drive three. Back it up to Time Machine on drive four. Use your two free 500GB partitions for other stuff.

I have a five disk setup on my G5 (two internal, three eSATA). For performance, my Photoshop scratch disk is a 32GB four disk stripped array. Each component is a 8GB partition off four of the five disks. Of course I would have better performance with dedicated drives, but this wastes much less space and is still very fast.

My MacPro setup is likely to be similar.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back