[zfs-discuss] Intel X25-M page size (Was: Request for assistance: Building drive identifier database)

Durval Menezes durval.menezes at gmail.com
Sat Sep 28 06:24:40 EDT 2013

Hi Gordan,

On Fri, Sep 27, 2013 at 7:33 PM, Gordan Bobic <gordan.bobic at gmail.com>wrote:

> On 09/27/2013 10:10 PM, Durval Menezes wrote:
>> Hi Richard,
>> On Mon, Aug 12, 2013 at 12:09 PM, Richard Yao <ryao at gentoo.org
>> <mailto:ryao at gentoo.org>> wrote:
>>     The INTEL SSDSA2M080G2GC is the 80GB X25-M G2. It uses 34-nm NAND
>> flash
>>     with a 4096-byte page size. [...]
>> I'm bringing up a new machine with ZFS and I put my old  SSDSA2M080G2GC
>> in it, to be used as ZIL/L2ARC.
>> Before going on with ZFS, I tested the above assumption by creating a
>> partition in it, first purposedly misaligned in respect to 4KB, then
>> perfectly aligned to it; the test results show me *very* little
>> difference between both situations:
>>  [...]
>> Am I missing something?
> You're not. SSDs nowdays handle wear leveling very well, and you have no
> way of knowing where a particular sector might end up anyway.

Humrmrm... I may be wrong, but I think just wear leveling would *not*
change page-alignment behavior on a SSD: after all, WL supposedly just
remaps writes to different pages...  If the data is misaligned, the SSD
would still be writing two pages instead of one, no?

Let the firmware do it's job (which it does pretty well on good SSDs, like
> Intels) and don't worry about it. It's something worth bearing in mind just
> in case you find yourself using substandard SSDs, but on Sandforce based
> ones it's unlikely to make any difference. For a start, I'm moderately sure
> that those throw away 0-filled blocks and don't store them, and the more
> recent ones compress and dedupe the data internally on a per-sector basis
> as well - so all bets are well and truly off on what the alignment might
> end up being on each block when all that is done.

I was aware that today's SSDs really push it in terms of data manipulation
(like the compressing/deduping you mention below), but this is a 4 year-old
model... I wasn't expecting it to be that "smart".

> Save your worrying about alignment for crap SSDs and cheap SD/CF/USB flash
> media.

Or better yet, save money to buy good SSDs and other stuff, as life is too
short to waste on being screwed by bad hardware... :-)


> Gordan
> To unsubscribe from this group and stop receiving emails from it, send an
> email to zfs-discuss+unsubscribe@**zfsonlinux.org<zfs-discuss%2Bunsubscribe at zfsonlinux.org>
> .

To unsubscribe from this group and stop receiving emails from it, send an email to zfs-discuss+unsubscribe at zfsonlinux.org.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.zfsonlinux.org/pipermail/zfs-discuss/attachments/20130928/a1389861/attachment.html>

More information about the zfs-discuss mailing list