 |
 |
On 21 Jul 2011, at 15:22, Adam Davison wrote: > Hi, > > We're setting up our CREAM CE to interact with a central university run cluster, which uses SGE if that makes any difference. > > We've been assigned some slightly random set of non-sequential UIDs for our pool accounts. Does anyone know if you need to use the same UIDs for the pool accounts on the CE or does it make no difference? It's a lot easier if they are the same; and are often reuired to be. The LRMS (a.k.a. batch system) has to move output data files from the worker node back to the CE. This is normally done via rcp/scp and host.equiv type mechanisms, which require the UID's to be the same on the worker nodes and the CE's. It is possible to set up a cluster to use other mechanisms; but most of them use the same UID matching. To avoid UID matching, you need a way to move files in an authenticated manner, and apply ACL's, other wise one users jobs could access/delete data belonging to another user. Given that POSIX does that, if you match UID's, it's the easy way. In short: I'd match them, even if it's a hassle.