On Mon, Nov 30, 2015 at 7:47 PM, Timofey Titovets <nefelim4ag@xxxxxxxxx> wrote:
>
> On 30 Nov 2015 21:19, "Ilya Dryomov" <idryomov@xxxxxxxxx> wrote:
>>
>> On Mon, Nov 30, 2015 at 7:17 PM, Timofey Titovets <nefelim4ag@xxxxxxxxx>
>> wrote:
>> > Hi list,
>> > Short:
>> > i just want ask, why i can't do:
>> > echo 129 > /sys/class/block/rbdX/queue/nr_requests
>> >
>> > i.e. why i can't set value greater then 128?
>> > Why such a restriction?
>> >
>> > Long:
>> > Usage example:
>> > i have slow CEPH HDD based storage and i want it export by iSCSI proxy
>> > machine for ESXi cluster
>> >
>> > If i have so low queue depth, i just worry to get a performance
>> > problem because cluster can process request, but HV/VM can't send new
>> > request if queue in full state
>> >
>> > Then i have two solutions make several RBD devices for exporting or
>> > increase queue depth.
>> >
>> > What you think about it?
>>
>> Which kernel is that?
>>
>> Thanks,
>>
>> Ilya
>
> 4.2.5 on archlinux,
> Thanks
On 4.2 and above you can set queue depth at map time:
# rbd map -o queue_depth=129 ...
The reasons lie in the block layer, which has to support both single
queue drivers and drivers that are using blk-mq interface. There are
data structures that have to be preallocated on device setup, etc.
nr_requests setting just doesn't work for blk-mq devices.
Thanks,
Ilya
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
