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 using external metadata actually
improves your performance versus internal metadata, you have underlying
performance problems to fix."
Have you been investigating this possibility?
Or is the improvement specific to the type of write load your server
endures?
Lionel.
[1]
http://fghaas.wordpress.com/2009/08/20/internal-metadata-and-why-we-recommend-it/
Le 27/02/2013 23:13, Arnold Krille a écrit :
My experience with an ssd for (external) meta-data says that imrovement
is quite a lot!
You won't get faster continous writes, that is still limited by the
hdd. But you get much faster random-writes and the reason is this:
- With internal meta-data on hdd, each write (or until each barrier)
is followed by a disk-seek to the end of the disk where the
meta-data lives followed by a seek back to where you are writing.
And then you mix random writes at random positions...
- With external meta-data on another hdd, your data-disk doesn't have
to seek to the end of the disk anymore, step one of improvement.
- With external meta-data on ssd, you are only left with the seeks
during your normal random writes.
With todays disks and normal usage (unless you are netflix or google),
the real speed-improvement your users see/feel is not faster throughput
but lower latency.
Of course, using internal meta-data with the whole partition on ssd
gives you the best performance, but not everyone can buy enough ssds to
create a mirrored 6TB array of ssd.
3x2TB hdd + 160G ssd (for meta-data and the fast-loving databases)
times two on the other hand is actually affordable...
As to the original authors question: There is a manpage about drbdmeta
which describes the options to dump and restore the meta-data of an
offline drbd. So the action will be:
- stop the drbd-volume
- dump the meta-data
- change the config to point the meta-data to the new place
- restore the meta-data
- restart the drbd-volume
- wait for sync (only incremental, not a full sync) and repeat with the
other node
I did that with several volumes when our ssds arrived. Test the steps
with a scrap-drbd-volume before doing the procedure on production-data
to be sure.
Have fun,
Arnold
_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user
_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user
