Hi Craig,
right! I had also post one mail in that thread.
My question was, if the whole step to "chooseleaf_vary_r 1" take the
same amount of time like the setting tunables to firefly.
The funny thing: I just decompile the crushmap to start with
chooseleaf_vary_r 4 and see, that after upgrade tonight the
chooseleaf_vary_r allready on 1!
# begin crush map
tunable choose_local_tries 0
tunable choose_local_fallback_tries 0
tunable choose_total_tries 50
tunable chooseleaf_descend_once 1
tunable chooseleaf_vary_r 1
...
ceph osd crush show-tunables -f json-pretty
{ "choose_local_tries": 0,
"choose_local_fallback_tries": 0,
"choose_total_tries": 50,
"chooseleaf_descend_once": 1,
"profile": "firefly",
"optimal_tunables": 1,
"legacy_tunables": 0,
"require_feature_tunables": 1,
"require_feature_tunables2": 1}
Udo
On 20.12.2014 17:53, Craig Lewis wrote:
> There was a tunables discussion on the ML a few months ago, with a lot
> of good suggestions. Sage gave some suggestions on rolling out (and
> rolling back) chooseleaf_vary_r changes. That reminds me... I
> intended to try those changes over the holidays...
>
>
> Found it; the subject was "ceph osd crush tunables optimal AND add new
> OSD at the same time".
>
>
> On Sat, Dec 20, 2014 at 3:26 AM, Udo Lembke <[email protected]
> <mailto:[email protected]>> wrote:
>
> Hi,
> for information for other cepher...
>
> I switched from unknown crush tunables to firefly and it's takes 6
> hour
> (30.853% degration) to finisched on our production-cluster (5
> Nodes, 60
> OSDs, 10GBE, 20% data used: pgmap v35678572: 3904 pgs, 4 pools, 21947
> GB data, 5489 kobjects).
>
> Should an "chooseleaf_vary_r 1" (from 0) take round about the same
> time
> to finished??
>
>
> Regards
>
> Udo
>
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
