On 10/24/2013 11:40 AM, matthew patton wrote:
I thought (and maybe I'm wrong) that a good chunk of reserved space was
essential to allow the drive to efficiently manage its read-modify-write cycles.
Correct. The industry ~standard of 7% (basically the difference between GiB and
GB) is woefully inadequate for any kind of steady write load. ALL enterprise
SSDs use north of 20% and and I've seen as high as 50%.
I must admit all my 250G SSDs are partitioned down to 200G.
It was my understanding that bcache was explicitly designed to fill erase block
sized
chunks sequentially and discard them in whole units,
negating the requirement for the drive to actually perform RMW cycles
RMW is an essential and inalienable of how an SSD works.
Well, yes... but. I'd have though if you don't give the drive a reason
to perform an RMW, then it would be a reasonable assumption that perhaps
it won't actually do one. Perhaps I give the firmware authors too much
credit.
Every manufacturer can use different page and erase block sizes. And much of
the time they don't publish the specs publicly. So while Kent may have gone to
deliberate length to optimize the way BCache does IO by using aligned, suitably
large chunks (eg. 128KB-512KB) he has zero control over what the firmware
decides to do.
This is essentially true, however making your storage bucket size big
enough to believe it holds at least 1 full erase block would be a
reasonable assumption. Oops.. ass-u-me..
BTW, did you undo the retarded disk label that Linux has used for decades which
is guaranteed to cause mis-aligned I/O? I expect BCache will start it's data
area at 1MB offset from where the device starts. But it can't do much to remedy
the problem if you didn't align the partition or LV you handed BCache
correctly to begin with.
I'm pretty sure all my drives are properly aligned. I learned that very
quickly when I started using the WD "advanced format" drives, except for
SSD's I align on 1M insteak of 4k.
I don't actually use bcache in production as when I did my last storage
upgrade I just could not get it reliable (well before it hit the
mainline kernel). I just keep tabs on it with the intention of using it
when it develops the ability to mirror writeback data. In the mean time
I'm just running on a RAID10 of 6 SSD's.
--
To unsubscribe from this list: send the line "unsubscribe linux-bcache" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
