Hi Wido,
Thank you for your helping.
I' m using network with 10Gbps speed for my Ceph Cluster.
and I want to optimize it to read/write images faster with object size=
16/32MB. Any your idea?
>> If you use the 'rbd' tool you can set the --order flag to set it to
anything else then the default 22 (4MB).
Also your reached way, Can I use arrgument configurations from .. and
don't nead to use rbd with --order ?
"rbd_default_stripe_count": "1",
"rbd_default_stripe_unit": "*4194304*",
to
"rbd_default_stripe_count": "1",
"rbd_default_stripe_unit": "*8388608*", #(16MB)
Thanks Wido!
--
Tuan
On 04/06/2014 04:34 AM, Wido den Hollander wrote:
On 04/05/2014 07:15 AM, Ta Ba Tuan wrote:
Hi everyone
My Ceph cluster is running, I'm plaining to tune my Ceph performance.
I want to increase object size from 4M to 16MB (maybe 32MB,..)
With the fomular: "stripe_unit" * "stripe_count" equals "object_size",
i'm thinking to change this following option:
"rbd_default_stripe_unit" from *4194304 => 8388608 (16MB)* and
"stripe_count" still keeping default
(ceph --admin-daemon /var/run/ceph/ceph-osd.74.asok config set
*rbd_default_stripe_unit **8388608* (on any osd(s))
Righ? Please guide me.
No, not completely. The striping/object size settings have to be set
when creating an RBD image.
If you use the 'rbd' tool you can set the --order flag to set it to
anything else then the default 22 (4MB).
The OSDs don't care about the object size, they simply serve objects.
Any reason why you want to go for 16 or 32MB? Just wondering. The 4MB
default seems to work fine, but I can imagine situations where a
larger object size works better.
Wido
Thank everyone!
--
Tuan
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
