Several important developments are underway on the cernsp service. Firstly the parc engineering service has been absorbed into it. For interactive work, parc users should login to interactive nodes as normal cernsp users. For batch work, either using loadleveller or their special ride command, they have eight dedicated SP2 nodes, and we have started to merge the previous eight-node SP1 system of parc as the seventh frame of the SP2. This has reduced the number of batch nodes for normal users, so, in compensation we have bought twelve new IBM 850 powerpc nodes and are integrating them into the cernsp service. These nodes are a different cpu chip but are fully binary compatible, except when special compiler architecture optimisation flags are used (e.g. hepf77 -qarch pwr2.) To allow for this, we will be introducing loadleveller parameters requiring work to execute on the right nodes. The new powerpc nodes do not connect to the high speed switch and will not perform well for data intensive work, so we are also introducing loadleveller parameters to allow jobs with low input/output requirements to be preferentially run on these new nodes. These will all be documented in the next release of the ``CERNSP Introductory User Guide''.
Finally we are planning to migrate the whole cernsp cluster from version 3.1 of the AIX operating system to version 4.1. We have already blocked off nine nodes and are running the new AIX on them with selected test users. We hope to migrate the whole cluster before the end of March in order to have a stable service before the accelerators restart. Please subscribe to the cern.cernsp newsgroup if you want to be kept informed about all these developments.