Alessandra Forti wrote:
> Hi Graeme,
>
> what do you mean low volume? low internet traffic? not cpu intensive?
> small disk space required? low maintainance? really high trust?
Sorry, "low volume" was probably an unhelpful term to use. All I was
wanting to do was draw a distinction between those services which can
run in a webservices/GRACE container (like a claims service, a message
service) and those services which couldn't (like xrootd).
For containerised services we can have a generic method for install, for
which I sketched a suggestion earlier; for the others I think that LCG
should provide packaged RPMs.
Many (perhaps most?) containerised services from LHC VOs will be low
traffic, not CPU intensive and required little disk, but not
necessarily. High trust is never given - that's why it's in a container.
(As an aside there is a notion that web services are inherently low
performace, which I don't think is borne out in practice. Well written
web services do perform well.)
>
> Although I haven't seen any of the experiments giving any explanation
> yet, I think that since we have started to ask loudly for explanations
> we have gained 2 things:
>
> 1) A well written policy
And it's a pretty decent looking policy now...
> 2) The most appalling requests have been dropped, I would say overnight.
It's not clear whether Atlas really did want root access (they certainly
never needed it). To be honest it read more like someone "thinking out
loud" on the wiki page.
>
> We still have to discuss the operational part
I think one of the things going for the GRACE scheme is that it's far
more "operationally trustworthy". Less opportunity for experiments to
make a mess of things - you can see all the RPM postinstall scripts, log
access to the web server, etc.
> and I still have massive
> reservations on the long term because I believe that once these machines
> are in place we will never get rid of them.
I think that's very likely.
> Nobody will do anything to
> insert the software as a general grid component.
It has to be a responsibility on the middleware developers and LCG team
to identify areas where components can be genaralised. It's clear that
it's not going to come from the experiments (nor, indeed, should it).
Cheers
Graeme
--
--------------------------------------------------------------------
Dr Graeme Stewart http://www.physics.gla.ac.uk/~graeme/
GridPP DM Wiki http://wiki.gridpp.ac.uk/wiki/Data_Management
|