Future Upgrades: SSD's - To RAID or not to RAID? Formats, Alignment, Etc.

[oshunluvr]

Other thread discussions led me to this topic: Myself and at least one other forum member (I'll allow him to "out" himself) is considering a 4 disk SSD RAID0 configuration - Why? Mostly just because.

I currently have 3x500gb WD Caviar Blacks upgraded to Enterprise RAID firmware, but a year or so back I switched from RAID0 to BTRFS-RAID0 out of curiousity and have kept it that way. I love the subvolume and backup/snapshot features of BTRFS.

My current hardware allows 6 sata drives. Since my DVD is IDE, I'm planning on keeping 2x500GB platter drives for backup storage and thus will have room for 4 SSD's.

My current plans are 4x Mushkin Enhanced Chronos Deluxe 60GB (currently $75 each on Newegg). This gives me 240GB total space which translates to 3 20GB install partitions, 8gb swap, and the remainder 172GB for my data partition as is my fashion. I have considered Patriot and OZC drives, but I like the smaller Mushkin company's use of better chips in their Deluxe drives (look for the -DX in the part #). I also want sata III because I will be swapping out mobo's next year and will have sata III interfaces then.

With the new SSD configuration, I will be testing a BTRFS install along side a RAID0/ext4 install both to see how they test against each other and also to see if there are any real daily-use differences between the two. When you add in auto-backup performance/ease-of-use along with other regular uses, this could show a difference.

I would like to hear others opinion/comments on my ideas as well as any real experience with these topics if anyone cares to share. I know one forum member is using BTRFS RAID on SSD's now and when we last discussed it was doing fine.

One first obvious thing to talk about is partition alignment and the role it plays in speed and longevity of SSD's. My research leads me to believe it is as simple as formatting using 32 heads/sectors and then start the first partition at the 2nd cylinder. Additional data alignment is required for LVM if used, but this also is a simple --dataalignment 512K switch when creating the pool.

[lcorken]

I'm curious about the Raid0 thing. Theoretically Raid0 should be nearly twice as fast with 2 drives relative to one. But I recall reading that some folks are disappointed in that 2 are only slightly better that one. ??

Ken.

[oshunluvr]

In my experience with RAID0 on platters (disc based hard drives) you do get a nearly double speed transfer rate, but no real reduction in overhead, i.e. seek and system time. This does indeed have huge benefits.

I have used RAID0 in a 4 disk array for five years and loved the massive speed increase. I even experimented with a RAID0 array over-laying a RAID1 array for true redundency but it was so slow as to be the same as no RAID at all so why bother?

I believe there is a potential for the SSD's to be so fast as to have a diminishing return rather rapidly. What I mean is, with 4 hard drives in RAID0 the speed difference is immediately noticable and very obvious. It "feels" four times faster. It will be interesting to see if 4 SSD's "feels" four times faster than a single SSD, since they're already very fast.

[SteveRiley]

The thread that spawned this one has a link to a Hardware Canucks test showing that a RAID-0 array of two SSDs performs approximately twice as fast as a single SSD. I was pleasantly surprised by that.

[oshunluvr]

So Steve, I gather you're mostly a Laptop/Tablet user. Any of those devices have a second drive bay?

[SteveRiley]

So Steve, I gather you're mostly a Laptop/Tablet user. Any of those devices have a second drive bay?

Exclusively. Haven't had a desktop in years.

My Lenovo T520 has an optical drive, which I could replace with an UltraBay containing a second SSD. But I carry so little data locally that I'm not sure having another drive would help. I am rarely without high-speed Internet connectivity. I use AWS for all my heavy-lifting storage and computational needs.

[dibl]

I'm intrigued by your project, oshunluvr -- I'll be very interested to learn how it goes. I've got a feeling -- can't quantify it -- that the 4-drive array on a SATA III bus may not be noticeably faster than a 2-drive array, but I'd be happy to be wrong about that.

Also, OCZ's FAQs say they still don't support trim when the drives are in RAID configuration. Is that also the case with Mushkin's SSDs? If true, it would seem like the performance will necessarily degrade over time -- I suppose it depends on how much/frequently you change the contents of the drives.

BTW, my btrfs filesystem, on a pair of WD 1002FAEX drives on a SATA III bus, continues to chug away with zero drama -- coming up on 18 months of continuous use now.

EDIT: Re:Alignment, I wrote this some time ago -- it is mostly still correct, AFAIK. The new versions of fdisk all start the first partition at 2048, so it will be aligned (at the expense of some wasted space) -- you need an older fdisk to set the heads and sectors now, because the new versions won't allow it.

[andystmartin]

An interesting project. I'll be interested in the results.

The downside of the speed gains of RAID0, though, is reliability. By striping across 4 volumes, theoretically, the reliability of the logical disk is divided by 4. The other aspect of SSDs (I've heard) is that they tend to fail catastrophically without warning. Whereas rotating disks often give hints that they are about to fail.

Nevertheless, this is a cool experiment.

[oshunluvr]

Agreed, however I am experienced enough with RAID0 to make backups. I usually set a rdiff-backup cron job and backup nightly.

Dibl: That was my thought too. The difference in "real world" use may not make the added complexity of a 4x SSD RAID worth the cost/trouble.
I am buying SATA III drives but I am currently limited to a SATA II interface. Doubtless, when I do upgrade the stats will change and the differences may be even less.


http://www.xbitlabs.com/articles/sto...ssd-raid0.html

The above article pretty much lays out most of what I had planned, with the exception of using 3 or 4 drives rather than just two. Since it appears SSD cannot trim in RAID configuration, I may stick with BTRFS (assuming it supports trim in a RAID format).

Possibly, I could use BTRFS for my home partition and RAID0 for the install, since it will be unlikely that I will fill the install partition prior to the next OS upgrade. I suppose with the amount of time required to do an install, I could just re-install whenever the performance degraded noticably.

[oshunluvr]

Dibl: How did you format your btrfs partition? I am planning -m raid1 -d raid0

What is your opinion on this?

EDIT: A bit more reading reveals there is no real reason to use RAID1 for the metadata in this configuration because you won't be able to recover anything anyway.

[dibl]

I followed the wiki and, since the default configuration seemed like what I wanted (mirrored metadata, striped data) I went with it.

mkfs.btrfs /dev/sdb /dev/sdc

where the 2 unformatted new drives were /dev/sdb and /dev/sdc.

I remember getting /etc/fstab set up correctly was a bit of a struggle, between "sub-UUIDs" and the correct way to list the two devices on the mount line. When I built this rig, compression was not yet released, so I have subsequently added that option.

Inasmuch as btrfs is really designed to support massive storage and server performance, and since it is not a speed demon compared to ext4, I'm skeptical about it as a good filesystem for your OS partition. Also, one caveat -- btrfs is a really bad filesystem choice for partition where you intend to keep a VM -- I think they're working on a fix for it, but it truly sucks for that particular application.

FYI, on my OCZ RevoDrive where I have my OS and a couple of VMs, I use ext4, with the journalling turned down to once every 120 seconds. It's a big desktop rig on a UPS, so I'm willing to go that far risking data loss due to power loss or something breaking.