On Wed, Jun 24, 2015 at 5:55 PM, Nick Fisk <[email protected]> wrote:
> That kernel probably has the bug where tcp_nodelay is not enabled. That is
> fixed in Kernel 4.0+, however also in 4.0 blk-mq was introduced which brings
> two other limitations:-
>
>
>
> 1. Max queue depth of 128
There will be an option in 4.2 to crank it up to ~10000.
>
> 2. IO’s sizes are restricted/split to 128kb
I'd advise using 4.1 on the kernel client boxes. The 128k limit is
blk-mq layer issue, it should be fixed in 4.2 as well, and unless your
workload is steady sequential direct I/Os, it's not going to matter
much.
Thanks,
Ilya
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
