SSD pool sanity check

[Frank Christopher via illumos-zfs]

I've seen hints of folks wanting to utilize SSD pools, but not a lot of
data on how it works. I'd like to experiment with doing the same on a
virtualization cluster with several thousand VMs. A primary goal of course
seeing if I can enable dedupe without the performance penalty experienced
with mechanical drives.


Wondering about protecting SSD endurance:


* Does it make sense to set recordsize to 4k or some value that would align
with the internal data structure in the SSD in order to avoid a stripe
re-write each time a VM performs a write?


* If I can put an low-latency SLOG device in place like a ddrdrive or
fusionio, would flash endurance possibly benefit from the write coalescing
or is this less of an issue if I use a fixed recordsize?



Thanks,
Frank



-------------------------------------------
illumos-zfs
Archives: https://www.listbox.com/member/archive/182191/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182191/23047029-187a0c8d
Modify Your Subscription: https://www.listbox.com/member/?member_id=23047029&id_secret=23047029-2e85923f
Powered by Listbox: http://www.listbox.com

[Bob Friesenhahn via illumos-zfs]

* If I can put an low-latency SLOG device in place like a ddrdrive or fusionio, would flash endurance possibly benefit from
the write coalescing or is this less of an issue if I use a fixed recordsize?

When synchronous writes are involved, then flash endurance and
performance would additionally benefit.

Ordinary SSDs still suffer from limited write transaction capacity.

It is best to keep pool devices unburdened as much as possible so that
they are immediately available to support reads.

Bob

[Matthew Ahrens via illumos-zfs]

On Thu, Jun 12, 2014 at 7:38 AM, Frank Christopher via illumos-zfs <

Post by Frank Christopher via illumos-zfs
I've seen hints of folks wanting to utilize SSD pools, but not a lot of
data on how it works. I'd like to experiment with doing the same on a
virtualization cluster with several thousand VMs. A primary goal of course
seeing if I can enable dedupe without the performance penalty experienced
with mechanical drives.
* Does it make sense to set recordsize to 4k or some value that would
align with the internal data structure in the SSD in order to avoid a
stripe re-write each time a VM performs a write?

The recordsize property has no impact on write alignment. Treating the SSD
as a 4KB-sector device would cause all i/o to the SSD to be 4KB-aligned.
This may help performance for some SSDs.

--matt

Post by Frank Christopher via illumos-zfs
* If I can put an low-latency SLOG device in place like a ddrdrive or
fusionio, would flash endurance possibly benefit from the write coalescing
or is this less of an issue if I use a fixed recordsize?
Thanks,
Frank
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify
<https://www.listbox.com/member/?&>
Your Subscription <http://www.listbox.com>

-------------------------------------------
illumos-zfs
Archives: https://www.listbox.com/member/archive/182191/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182191/23047029-187a0c8d
Modify Your Subscription: https://www.listbox.com/member/?member_id=23047029&id_secret=23047029-2e85923f
Powered by Listbox: http://www.listbox.com

[Frank Christopher via illumos-zfs]

Post by Matthew Ahrens via illumos-zfs
On Thu, Jun 12, 2014 at 7:38 AM, Frank Christopher via illumos-zfs <

Post by Frank Christopher via illumos-zfs
I've seen hints of folks wanting to utilize SSD pools, but not a lot of
data on how it works. I'd like to experiment with doing the same on a
virtualization cluster with several thousand VMs. A primary goal of course
seeing if I can enable dedupe without the performance penalty experienced
with mechanical drives.
* Does it make sense to set recordsize to 4k or some value that would
align with the internal data structure in the SSD in order to avoid a
stripe re-write each time a VM performs a write?

The recordsize property has no impact on write alignment. Treating the
SSD as a 4KB-sector device would cause all i/o to the SSD to be
4KB-aligned. This may help performance for some SSDs.

Correct me if I'm wrong, but if I use the default 128K recordsize and the
VMs start creating those larger records because it made sense upon
creation, but then they start (re)writing smaller chunks of that record...
I'm wondering if flash endurance would benefit from an initial 4K
recordsize by avoiding the rewrite of the entire record?


Thanks,
Frank



-------------------------------------------
illumos-zfs
Archives: https://www.listbox.com/member/archive/182191/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182191/23047029-187a0c8d
Modify Your Subscription: https://www.listbox.com/member/?member_id=23047029&id_secret=23047029-2e85923f
Powered by Listbox: http://www.listbox.com