Hi Neal,

Can you add user jchapman to LSF user group atlpetagrp? Also, when can we
enable LSF queue atlpetaq?

Hi Charlie,

For usage with direct login to SLAC machines, atlint01 is available (8 cores
and 8GB). Memfs machine can also be use via batch. I think Randy and Neal
want to know how long John will use them.

For "production" use, I don't know if you mean the normal atlas production
channel via Panda. If so, someone else in ATLAS production will have to
decide how to handle this large memory requirement. A pilot based system
doesn't really tells a site the job requirements because pilots don't know
what jobs they will run. The only thing SLAC can do is to setup different
Panda "sites" and imposed jobs requirement at site level. However in this
case setting up another Panda site will not help because normal production
will not use it.

If the "production" will be done by John himself, I think we can setup
another site ANALY_SITE_TEST (despite the prefix ANALY_) and have the site
use the dedicated atlaspetaq.

Regards,
Wei Yang  |  [log in to unmask]  |  650-926-3338(O)

> From: "Young, Charles C." <[log in to unmask]>
> Date: Thu, 12 Mar 2009 08:26:27 -0700
> To: unix-admin <[log in to unmask]>
> Cc: atlas-sccs-planning-l <[log in to unmask]>, "Moss,
> Leonard J." <[log in to unmask]>
> Subject: RE: [SLAC #163429] Request to use few memfs machines for ATLAS
> testing
> 
> Hi Wei,
> 
> I know John has run the jobs (that require ~4 GB) at CERN already. There are
> no fundamental problems. However, it would be useful to make a quick test
> here, either interactively or normal batch. There is enough memory on the
> typical interactive node, but it may impact other users on that node. Hence
> the suggestion of doing it in a fenced off area. Once production starts, there
> may be unforeseen problems and it would once again be useful to be able to go
> in and debug. 
> 
> When it comes to production, we will want to go through the normal channels as
> much as possible. The only things special that I am aware of is the large
> memory requirement. If that means defining another "site", I guess we have to
> do it. Does this mean that each "site" is expected to be homogeneous with no
> variations in the properties of its CPUs? Memory, swap space, etc. I would
> naively expect a bit more flexibility. Cheers.