We did some CBT tests here a few months ago which included some dmcrypt
comparisons - the performance hit was non-zero, but close enough, around
~2-3%.
(CentOS 7.2 with E5-2630 v4 cpus, jewel release, default dmcrypt
parameters which IIRC is AES-XTS now).
G.
On 1/3/2017 7:48 AM, Adrien Gillard wrote:
There has been talks on the subject in the mailing list before [1] which
concur with Nick's experience as long as you use AES-XTS.
[1] http://lists.ceph.com/pipermail/ceph-users-ceph.com/2016-March/008444.html
On Tue, Jan 3, 2017 at 2:30 PM, Nick Fisk <nick@xxxxxxxxxx
<mailto:nick@xxxxxxxxxx>> wrote:
__ __
__ __
*From:*ceph-users [mailto:ceph-users-bounces@xxxxxxxxxxxxxx
<mailto:ceph-users-bounces@xxxxxxxxxxxxxx>] *On Behalf Of *Kent Borg
*Sent:* 03 January 2017 12:47
*To:* M Ranga Swami Reddy <swamireddy@xxxxxxxxx
<mailto:swamireddy@xxxxxxxxx>>
*Cc:* ceph-users <ceph-users@xxxxxxxxxxxxxx
<mailto:ceph-users@xxxxxxxxxxxxxx>>
*Subject:* Re: performance with/without dmcrypt OSD____
__ __
On 01/03/2017 06:42 AM, M Ranga Swami Reddy wrote:
____
On Tue, Jan 3, 2017 at 6:17 AM, Kent Borg <kentborg@xxxxxxxx
<mailto:kentborg@xxxxxxxx>> wrote:____
__ __
Assuming I am understanding the question...
If there isn't too big a performance hit, it makes disk
disposal (we expect disks to die, right?) much simpler.____
__ __
__ __
OK. Thanks. But if I have a big volumes in TB size (10 TB
volume) and writing/reading from the big volumes - will impact
on performance like write and read speed?____
I'd like to know, too.
-kb____
__ __
Not specifically related to Ceph, but I built a 14 disk RAID 6 array
(mdadm) for a recent “secure high performance seeding device in a
briefcase” project and used dmcrypt on it. I could easily obtain
over 1GB/s reads and writes. From tests there was no noticeable
performance impact and CPU usage on a Xeon E3 was nothing to be
concerned about. All modern CPU’s will HW accelerate the process if
you use the AES-XTS cipher, I suspect there might be a severe
performance impact without.____
__ __
Also as Ceph+network itself brings a fair amount of overhead, I
wouldn’t suspect that dmcrypt would introduce any noticeable
overhead of its own.____
--
Graham Allan
Minnesota Supercomputing Institute - gta@xxxxxxx
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
