This is a broad question -- I'm not asking about a particular RAID level or OS.
-
I'll guess a qualified Yes. As all software raid on the windows side treats the drive as a single drive. Only the volume is treated like RAID.
You would have to be sure to plug the drive into a drive controller that doesn't have BMC or RAID support. Essentially if speedfan, hdtune, and such can see SMART data I would expect the Trim command to make it to the drive.
Windows 7 and Windows Server 2008 R2 support Trim. That would mean software RAID 0, 1, or 5 could support trim if I'm right.
That leaves Server 2008 32bit, Server 2008 64 bit (not R2), Server 2003 (all versions) out on this. Doubly so for Vista and XP which can't create software RAID volumes.
Unfortunately on the Windows side of things RAID 10 is not supported as a fully software based raid level. You would have to do RAID 0 or 1 in hardware and then stripe or mirror as needed on the software side to get the 1+0 or 0+1 array you truly wanted and at that point no TRIM support would apply just as if you did RAID 10 entirely in hardware.
In addition many hardware raid controllers will be slower than software RAID with SSDs. See http://it.anandtech.com/IT/showdoc.aspx?i=3532&p=9 for benchmarks showing this concept.
Massimo : Why the downvote?!?Helvick : I see your point but do you have any documentation to back this up ?TRIM support in a RAID setup requires that the file deletes (which happen at the volume level) are communicated to the individual drives via their drivers. For this to be true the Windows Software RAID layer will have to handle that communication, simply supporting TRIM at the filesystem and drive level isn't enough - I haven't found any evidence that this happens.pplrppl : Helvick. I don't have documentation. I'm not 100% sure. Since Microsoft doesn't clearly state one way or the other that I know off the only way to tell would be to run bechmarks on a pair of trim supported SSDs using software RAID 0 or 1 that isn't the boot partition on a W7 or Server 2008 R2 box. It's no small amount of work to document this behaviour and prove that it would or would not work as my hypothesis states. I'd be happy if someone would prove me right or wrong with testing and/or documentation. http://www.anandtech.com/storage/showdoc.aspx?i=3667&p=2 mentions a rough test method.jimmygee : This can be proved fairly simply. Use this tool to see if the OS can communicate with the SSDs firmware and return affirmative TRIM support: http://crystalmark.info/software/CrystalDiskInfo/index-e.html Then use the manual verification technique described in this answer to see if it _actually_ works: http://serverfault.com/questions/94017/ssd-tweaks-recommended-configuration/127558#127558jimmygee : Apparently the latest update to the Intel Matrix RAID system will pass TRIM commands: http://www.hardwarecanucks.com/news/cpu/intel-chipset-driver-brings-trim-support-raid-setups/Fantius : But only to a single SSD, not to the array.pplrppl : @fantius, each SSD is visible to the OS in software RAID so yes, it could be passed to all drives in the array. That is the point about software RAID, only in "fake raid" or hardware based raid are the drives hidden and presented to the OS as a lump sum.From pplrppl -
Not yet, no, at least as far as I'm aware.
The issue is the disconnect between the filesystem layer and the LVM systems used to handle the software RAID - generally they abstract away the type of device from the filesystem so it doesn't know to use TRIM.
The first people to support it in software will probably with Sun with ZFS, which doesn't have a traditional LVM layer but instead operates as a unified pool. They're already working on TRIM support being added to ZFS, when it happens I think it'll work across all the ZFS functionality including software RAID.
From Ewan Leith
0 comments:
Post a Comment