On Thu, 6 Mar 2003, Ian Stokes-Rees wrote:
> It seems to me to be a problem if single user jobs are dumping 500+ megs in
> /home and then just leaving them there.
>
> Do job outputs "hang around" in the particular /home output directories
> until they are retrieved by a user?
I think the sandbox files are sent back to the RB and deleted when the
job finishes (they *do* hang around on the RB, hence they should only
be small). However, I think there is nothing which prevents a job putting
files in the home directory and leaving them, it's just bad practice.
> Does this not suggest that there is going to be an aweful lot of NFS traffic
> between wherever /home is commonly mounted and the particular WNs? Does
> this also not mean that there will be double NFS traffic from a local SE to
> the /home machine and then on to the WN?
Yes, at least with a big cluster. With the current implementation it's
almost certainly more efficient to read and write files on the SE
directly at many sites, but users don't seem to want to do it. As Steve
said you can make the WNs use /tmp which is a bit better (as long as the
local disks are big enough).
Stephen
|