Re: autosave/restore software fails

[Tim Mooney <mooney@aps.anl.gov>]

D. Peter Siddons wrote:
> Hi all,
>    We have been using a host-based tool from Los Alamos (Bob Dalesio is 
> the author) called ss-arch, which seems to have disappeared from the 
> distribution. It uses CA to maintain the restore file. At least it 
> avoids the NFS vagaries. For a beamline the number of really critical 
> variables is modest enough that this approach can work. I don't know the 
> SLAC tool.
> Pete.

For a large number of variables whose values rarely change, a host-based
save tool can have a significant performance advantage over a crate-based
tool.  One shortcoming in autosave's model is that the entire .sav file
has to be rewritten if any PV in the list changes, and some crates running
autosave are saving over 6000 PV's.

There's no fundamental reason the save part of autosave could not run on
a host.  I like it on the crate because, there, it's most likely actually
to be running whenever the crate is running, but the save part is a pretty
vanilla-looking CA client, and porting it to run stand-alone on a host
would be easy.  Plus, most operating systems allow you to do things like
rename files, and I'm sure some actually flush when you call fflush(),
so there could be significant simplifications as well.

-- 
Tim Mooney (mooney@aps.anl.gov) (630)252-5417
Beamline Controls & Data Acquisition Group
Advanced Photon Source, Argonne National Lab