Same observations here. Large linear writes on a single-spinde backing
device are throtteled to less than 50% of their non-DRBD transfer rate
when using internal metadata. After moving the metadata to a SSD, almost
80-90% of the orginal write rate will be reached again (which is a
performance g
On Thu, Feb 28, 2013 at 09:49:16AM +0100, Lionel Sausin wrote:
> Dear Arnold,
>
> Thanks for your feedback.
> It's interesting because, normally, writes do not directly translate
> to head seeks (thanks to dirty pages, caches, NCQ, firmware-level
> optimization...), and ideally barriers should be
On 02/28/2013 12:22 PM, Arnold Krille wrote:
>> It's interesting because, normally, writes do not directly translate
>> > to head seeks (thanks to dirty pages, caches, NCQ, firmware-level
>> > optimization...), and ideally barriers should be disabled (and caches
>> > reliable).
> If you want secu
On Thu, 28 Feb 2013 09:49:16 +0100 Lionel Sausin
wrote:
> It's interesting because, normally, writes do not directly translate
> to head seeks (thanks to dirty pages, caches, NCQ, firmware-level
> optimization...), and ideally barriers should be disabled (and caches
> reliable).
If you want sec
Dear Arnold,
Thanks for your feedback.
It's interesting because, normally, writes do not directly translate to
head seeks (thanks to dirty pages, caches, NCQ, firmware-level
optimization...), and ideally barriers should be disabled (and caches
reliable).
Florian Haas once suggested[1] that "if
>
> I wouldn't expect anything like the gains of bcache/flashcache/enhencio.
> Normally internal metadata are just as fast, thanks to the write cache
> of your disks and RAID adapter. Those are much faster than SSDs and
> metadata are small enough.
> However you may benefit from external metadata
Lionel Sausin wrote:
>I wouldn't expect anything like the gains of
>bcache/flashcache/enhencio.
>Normally internal metadata are just as fast, thanks to the write cache
>of your disks and RAID adapter. Those are much faster than SSDs and
>metadata are small enough.
>However you may benefit from
On Wed, 27 Feb 2013 18:32:07 +0100 Lionel Sausin
wrote:
> I wouldn't expect anything like the gains of
> bcache/flashcache/enhencio. Normally internal metadata are just as
> fast, thanks to the write cache of your disks and RAID adapter. Those
> are much faster than SSDs and metadata are small eno
I wouldn't expect anything like the gains of bcache/flashcache/enhencio.
Normally internal metadata are just as fast, thanks to the write cache
of your disks and RAID adapter. Those are much faster than SSDs and
metadata are small enough.
However you may benefit from external metadata when your
Smallish SSD's are dirt cheap these days and I have a spare slot in my servers
so I was thinking of putting one in and running bcache (layering would look
like drbd->raid0->bcache->sd[ab]). It's a bit of mucking around to set up as
bcache isn't in the kernel at this time and is targeted against
10 matches
Mail list logo
