[sheepdog] [PATCH 0/5] add graph stat function

Liu Yuan namei.unix at gmail.com
Sat Aug 10 17:10:30 CEST 2013


On Sun, Aug 11, 2013 at 12:06:02AM +0900, MORITA Kazutaka wrote:
> At Sat, 10 Aug 2013 22:21:06 +0800,
> Liu Yuan wrote:
> > 
> > This patch set add a new command 'graph stat', this is useful to get how much
> > time each function spends in a top style view.
> > 
> > For a simple glance, we can see that our event_loop is doing too much.
> > 
> > $ collie trace graph stat
> >    Function Name                    |  Time(ms)
> > --------------------------------------------------
> > eventfd_xread                       | 110817.996257
> > do_process_work                     | 87415.051934
> > do_push_object                      | 84434.906931
> > push_cache_object                   | 84426.040582
> > gateway_create_and_write_obj        | 82382.884469
> > exec_local_req                      | 82038.293182
> > gateway_forward_request             | 81968.317720
> > event_loop                          | 48661.499542
> > do_event_loop                       | 48650.993934
> > xpwrite                             | 38091.275768
> > sheep_do_op_work                    | 36608.988329
> > peer_create_and_write_obj           | 36608.802587
> > default_create_and_write            | 36607.231493
> > object_cache_push                   | 28783.994126
> > do_background_push                  | 27332.938972
> > local_release_vdi                   | 3455.208848
> > object_cache_flush_vdi              | 3452.805846
> > read_cache_object_noupdate          | 2309.550495
> > ...
> 
> Thanks, I think of trying to implement the similar feature.  How about
> printing the number of calls for each function, too?  I think it would
> help us to profile a bottleneck.

Looks good to me, I'll include nr of calls in v2

Thanks
Yuan


More information about the sheepdog mailing list