Re: [OpenAFS] appropriate -cb numbers for fileservers?

2022-09-08 Thread Benjamin Kaduk 
On Thu, Sep 08, 2022 at 10:00:44AM -0400, Dave Botsch wrote:
> What's the best way to determine appropriate -cb numbers for
> fileservers? And is there a max that is "bad" like with -p ?
> 
> Note I'm still on 1.6 servers at the moment, which might change the
> answers to the above.

As Mark said, set it bigger than what your actual observed load is :)

It's interesting that the discussion at
https://gerrit.openafs.org/#/c/11629/ did not actually say much about the
callbacks value (though note that that change is still a work in progress).

-Ben
___
OpenAFS-info mailing list
OpenAFS-info@openafs.org
https://lists.openafs.org/mailman/listinfo/openafs-info

Re: [OpenAFS] appropriate -cb numbers for fileservers?

2022-09-08 Thread Mark Vitale 



> On 8 Sep 2022, at 10:00 AM, Dave Botsch  wrote:
> 
> What's the best way to determine appropriate -cb numbers for
> fileservers? And is there a max that is "bad" like with -p ?

Start with checking your current callback stats:

18:12:16 mvs1:sna-openafs-master mvitale$ xstat_fs_test mvs1 -onceonly -collID 3

Starting up the xstat_fs service, no debugging, one-shot operation


34 DeleteFiles
 0 DeleteCallBacks
   870 BreakCallBacks
   480 AddCallBack
 0 GotSomeSpaces
 3 DeleteAllCallBacks
 0 nFEs
 0 nCBs
 6 nblks
 8 CBsTimedOut
 0 nbreakers
 0 GSS1
 0 GSS2
 0 GSS3
 0 GSS4
 0 GSS5

The 'nblks' value is identical to the -cb value being used by the fileserver.
The 'nFEs' and 'nCBs' values should stay _well_ below 'nblks'.
So the best way to ensure that is to monitor these values periodically and
keep the data in a time-series database so you can chart them as needed.

Whenever either nFEs or nCBs exceeds nblks, you will see the 
'GotSomeSpaces' counter increment.  (The GSS* counters will also increment
in a way that is useful only to developers.)  Even one GotSomeSpaces event
is extremely bad for fileserver performance.

My rule of thumb is that when you see nCBs or nFEs approach 50% of nblks,
it is time to double your current value of the fileserver -cb option.

If you find that callback space is being consumed at an unreasonable rate,
there are additional (but very obscure) ways to identify which client(s)
are responsible for the callback load.  Let me know if you want more 
details on that.


Hope that helps,
--
Mark Vitale
Sine Nomine Associates

___
OpenAFS-info mailing list
OpenAFS-info@openafs.org
https://lists.openafs.org/mailman/listinfo/openafs-info