http://www.fusionio.com/Products.aspx Looks like a cool SSD to go with ZFS Has anybody tried ZFS with Fusion-IO storage? For that matter even with Solaris? -Jignesh -- Jignesh Shah http://blogs.sun.com/jkshah Sun Microsystems,Inc http://sun.com/postgresql
Last time I played with one of those, the problem was that it didn''t have any drivers for Solaris. It''s a PCIE device unlike something like the gigabyte I-Ram or Intel SSD or something. -- This message posted from opensolaris.org
I agree, it looks like it would be perfect, but unfortunately without Solaris drivers it''s pretty much a non starter. That hasn''t stopped me pestering Fusion-IO wherever I can though to see if they are willing to develop Solaris drivers, almost everywhere I''ve seen these reviewed there have been comments about how good they would be for ZFS. -- This message posted from opensolaris.org
zfs-discuss-bounces at opensolaris.org wrote on 09/24/2008 05:54:45 AM:> I agree, it looks like it would be perfect, but unfortunately > without Solaris drivers it''s pretty much a non starter. > > That hasn''t stopped me pestering Fusion-IO wherever I can though to > see if they are willing to develop Solaris drivers, almost > everywhere I''ve seen these reviewed there have been comments about > how good they would be for ZFS.Maybe you should stop pestering, as I just received a cold call from them made possible by them harvesting info on this list. -Wade
Regarding Solaris, or just a generic cold call? Either way it''s interesting to hear that they''re making calls. The impression I''ve had from all the press is that they''ve been struggling to meet demand. -- This message posted from opensolaris.org
Yes, we''ve been pleasantly surprised by the demand. But, that doesn''t mean we''re not anxious to expand our ability to address such an important market as OpenSolaris and ZFS. We''re actively working on OpenSolaris drivers. We don''t expect it to take long - I''ll keep you posted. -David Flynn CTO Fusion-io dflynn at fusion-io.com -- This message posted from opensolaris.org
Wohoo! Best news I''ve heard all week. Thanks for posting David :) -- This message posted from opensolaris.org
Just a thought, will we be able to split the ioDrive into slices and use it simultaneously as a ZIL and slog device? 5GB of write cache and 75GB of read cache sounds to me like a nice way to use the 80GB model. -- This message posted from opensolaris.org
D''oh, meant ZIL / slog and L2ARC device. Must have posted that before my early morning cuppa! -- This message posted from opensolaris.org
> Yes, we''ve been pleasantly surprised by the demand. > But, that doesn''t mean we''re not anxious to expand > our ability to address such an important market as > OpenSolaris and ZFS. > > We''re actively working on OpenSolaris drivers. We > don''t expect it to take long - I''ll keep you posted. > > -David Flynn > > CTO Fusion-io > dflynn at fusion-io.comThere is a fusion-io user here claiming that the performance drops 90% after the device has written to capacity once. Does fusion-io have a response? http://forums.storagereview.net/index.php?s=&showtopic=27190&view=findpost&p=253758 -- This message posted from opensolaris.org
On Tue, 21 Oct 2008, MC wrote:> > There is a fusion-io user here claiming that the performance drops > 90% after the device has written to capacity once. Does fusion-ioIsn''t this what is expected from FLASH-based SSD devices? They have to erase before they can write. The vendor is kind enough to deliver a pre-erased drive so the user feels an initial rush of exhilaration and feels good about the purchase. With standard hard drive interfaces the drive does not know if written data is useful or not so it can only erase when an overwrite is requested. This is because with hard drives, there is no need to "free" any data on the media. Usually the SSD''s erase block size is larger than the filesystem block size so more data needs to be erased than will be written, so some pre-existing data needs to be read and restored (potentially placing it at risk). With a higher level interface (e.g. to ZFS) then freed regions can be erased once they are added to the filesystem free list and will hopefully be erased by the time they are used again. It seems that optical media often support an erase mechansim so perhaps there is something in the IDE/ATA/SATA/SCSI/SAS protocols which can be used to explicitly erase blocks. Bob =====================================Bob Friesenhahn bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
Fascinating link, thanks for posting it. I was writing a nice long reply about what this means for usage, but I''ve just been on the Fusion-io web site, and it appears they have updated their documentation. They now state: Write MB/s: 600 Read MB/s: 700 Read IOPS: 102,000 (sustained 4k random reads) Write IOPS: 101,000 (sustained 4k random writes) It doesn''t look like I saved any early spec sheets, but I certainly don''t remember them saying anything about a sustained 4k iops rate before. That makes me think there''s a fair bit of truth to what was said. Since both IBM & HP have made announcements that they are working with Fusion-io, and a lot of Fortune 100 companies signed up to the technology too, it will be interesting to see how this plays out. -- This message posted from opensolaris.org