3.5in SAS flash drive for R900?

John LLOYD jal at mdacorporation.com
Fri Sep 10 18:58:21 CDT 2010


> Message: 2
> Date: Fri, 10 Sep 2010 10:53:44 +0100
> From: Tim Small <tim at seoss.co.uk>
...snip...
> This is basically due to the current generation of SSDs pretending a
> hard disk - when really they are flash (which has some very different
> physical properties).  Probably the best solution from an engineering
> point of view would be to use a flash-file-system (Linux has several -
> and they are designed to suite the physical properties of flash
> storage), and have the SSDs expose the raw underlying flash.  This
> however, is not the route the industry is taking....  TRIM is a piece
> of
> gaffer/duct-tape to fix this problem by providing a mechanism for the
> OS
> to tell the SSD which logical blocks it no-longer needs to work to
> preserve....
> 
> Without TRIM a workaround is to periodically discard all data by doing
> an ATA security-erase-unit, but this might not fit in with your
> anticipated usage.
> 
> bcache also looks very interesting, but is currently alpha-quality
> http://bcache.evilpiepirate.org/
> 
> BTW, the machine I'm using currently uses TRIM with both ext4, and
> btrfs
> on an Intel SSD, with an AHCI controller (Intel ICH10).
> 
> Tim.

Interesting.  Good to hear from someone who actually uses the stuff.

To avoid disk-drive-packaging and RAID-controller hassles, I'm thinking of an OCZ Z-Drive unit, like a p84 or p88 (4-way and 8-way stripes).  It's a plug-in PCI-e card.

Our application is write a big (100+GB) file and read it, once or twice, then delete the file and do it again.  With 2 to 4 concurrent threads.    Each thread is asynch to the others (so reading and writing are concurrent).  With magnetic disk it is truly a pain to get linear read-write trade-off (that is, read rate + write rate = a constant, across all possible rates.)  

So for now I'm thinking TRIM doesn't matter. (we have SUSE SLES10 SP3 which is only kernel level 2.6.16 plus some backports.)  I see XFS is supporting TRIM only recently so TRIM is not an option at the moment anyway.  We can run that trim script too.  And find out from OCZ if their controller supports it.


--John









More information about the Linux-PowerEdge mailing list