Hello Robert,

Le 08. 10. 15 13:54, Robert Heinzmann a écrit :
> Hi,
>
>> I think you did a really nice investigation work :-)
> Thanks ... was fun :)
>
>> I think that your patch is almost right, and we need to figure if we want to 
>> add specific directives for that. I tend to believe right now that with or 
>> without the POSIX call, the os will cache the file, and it sounds better to 
>> do it in advance, let see what other people are thinking about your idea.
>
> the point is, does it cache the file and evict the cache after the backup of 
> the file (what is with posix call)
>
> Example WITH fix:
> ----------------------
>
>    Server: 8 GB unused memory
>    Running Application (aka workload) = 4 of working set and cached files
>    Bacula backing up 800 x 100 MB FILE (8 GB in total)
>
> => After backup
>
> Running Application:
>    If bacula backs up the working set data: 0 GB cached
>    If bacula backs up other sets of data: 4 GB cached (e.g. a Snapshot of the 
> dataset which also has new filehandles) <=== Our case !!!!
> Bacula 100MB of cached (and never reused) data
>
> OR
>
> does bacula polute the whole file system cache
>
> Example WITHOUT fix:
> -----------------------------
>
>    Server: 8 GB unused memory
>    Running Application (aka workload) = 4 of working set and cached files
>    Bacula backing up 800 x 100 MB FILE (8 GB in total)
>
> => After backup
>
> Running Application: 0 GB cached
> Bacula 8 GB of cached (and never reused) data
>
>

I think that if I understand correctly what you wrote, the "fix" is 
better in all cases (and I agree with you), so I'm not convinced that
we should have a directive.

Best Regards,
Eric

------------------------------------------------------------------------------
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users