On Sun, Feb 28, 2010 at 06:26:31PM -0500, Jeff Moyer wrote:

If anyone is interested in a talk on SSDs, chime in. I can present my experiences with these devices. I've evaluated the performance of several vendors' products, ranging from the poor to the exemplary. I can talk about the modifications made in the block layer to accommodate for these devices, and the current implementation of the block discard support (which maps to ATA TRIM, which is important for device lognevity and performance) and how that is wired up to file systems such as ext4.

I'm interested. I was considering using SSDs for my boot drive. Is RAID protection necessary/advisable using SSDs?

On Mon, Mar 1, 2010 at 4:21 PM, Chuck Anderson wrote:

I'm interested. I was considering using SSDs for my boot drive. Is RAID protection necessary/advisable using SSDs? _______________________________________________ Wlug mailing list Wlug@mail.wlug.org http://mail.wlug.org/mailman/listinfo/wlug

SSDs have much higher reliability than HDD. Mirroring is the only RAID level that does not impose a write penalty. Mirroring SSDs is expensive Based on these 3 points, I'd say a backup is a much better solution than RAID, unless you want to (R)AID-0 them into a stripe for even more insane performance.

Chuck Anderson writes:

On Sun, Feb 28, 2010 at 06:26:31PM -0500, Jeff Moyer wrote:

...

If anyone is interested in a talk on SSDs, chime in. I can present my experiences with these devices. I've evaluated the performance of several vendors' products, ranging from the poor to the exemplary. I can talk about the modifications made in the block layer to accommodate for these devices, and the current implementation of the block discard support (which maps to ATA TRIM, which is important for device lognevity and performance) and how that is wired up to file systems such as ext4.

I'm interested. I was considering using SSDs for my boot drive. Is RAID protection necessary/advisable using SSDs?

The answer is, as always, it depends. I'll cover this in more depth when I give the talk, but any RAID algorithm that needs to initialize every block of the device is going to hurt performance on SSDs. There is some question about the durability/reliability of the drives, and some of those concerns are valid, others aren't. If you are diligent about backups, or if the disk just holds the OS, I wouldn't worry too much about RAID. HTH, Jeff

Chuck Anderson writes:

On Mon, Mar 01, 2010 at 04:46:44PM -0500, Jim Dibb wrote:

...

On Mon, Mar 1, 2010 at 4:21 PM, Chuck Anderson wrote:

...

I'm interested. I was considering using SSDs for my boot drive. Is RAID protection necessary/advisable using SSDs?

SSDs have much higher reliability than HDD. Mirroring is the only RAID level that does not impose a write penalty. Mirroring SSDs is expensive

Based on these 3 points, I'd say a backup is a much better solution than RAID, unless you want to (R)AID-0 them into a stripe for even more insane performance.

But then again, you didn't state how important your uptime of your system is. Sorry for leaving that bit out....

Uptime/availability isn't that important, but backups are since RAID *is* my backup (plus rsync to other RAIDed systems for the most important data). With traditional disk drives, it hasn't been too expensive to slap a 2nd system drive in for RAID 1/mirroring. SSDs are much more expensive, so if as you say they are more reliable, perhaps I can forgo the expense of doing RAID on them.

I've seen many claims that SSDs are more reliable than traditional HDDs, but I haven't seen data to back up those claims. My main concern is that SSDs are still young, so they haven't had much soak time. Whether or not the flash is reliable is only a small part of the battle. Buggy firmware is a much more likely problem at this stage in the game. No single point of failure is a good design point when dealing with your data. ;-) I remain cautiously optimistic (though I've already bricked one drive). Cheers, Jeff

"John Stoffel" writes:

...

...

...

...

...

"Jeff" == Jeff Moyer writes: Jeff> I've seen many claims that SSDs are more reliable than Jeff> traditional HDDs, but I haven't seen data to back up those Jeff> claims.

Some of the arguements I've read on SSD reliability are quite interesting. If you do the math on how much data you need to write to destroy an SSD with proper (note!) wear levelling, it's really quite high. But I think the real key is to write data to them in blocks of 512 bytes as much as possible, and to not write smaller amounts if at all possible.

Did you mean to say 512KB? You can't write smaller than the exported block size (which will be either 512 bytes or 4k). Perhaps what you meant to say was that you should try to issue writes that are aligned on and multiples of the erase block size.

Jeff> My main concern is that SSDs are still young, so they haven't Jeff> had much soak time. Whether or not the flash is reliable is Jeff> only a small part of the battle. Buggy firmware is a much more Jeff> likely problem at this stage in the game. No single point of Jeff> failure is a good design point when dealing with your data. ;-)

I think this is a key issue, the firmware and controllers, not the flash chips themselves.

Jeff> I remain cautiously optimistic (though I've already bricked one Jeff> drive).

Details?

During TRIM testing, we managed to get an OCZ Vertex to spew I/O errors, and then finally fail to be recognized by the BIOS upon reboot. Note that this isn't a normal workload (the drive is being pounded on) and TRIM is still very new. So, I am not saying that the drives are not reliable in general. I simply don't know. Cheers, Jeff