Quick summary: new silo1 and macomp00, new Matlab queues, new MySQL server, new CVS and git servers
Having reached the start of the summer term, here's an update on Maths research computing news:
New silo1 server and new silo mount layout: silo1 has replaced the existing silo server; silo users had already been informed of this individually by email and in addition, those who accessied silo directly rather than through its NFS shares were sent details of their new login accounts on silo1. silo1 is very similar to silo2 with 10.4 TB of storage, nearly three times that of silo.
At the same time the opportunity was taken to change the layout of both silo1 and silo2's mounts on NFS client systems to a simpler and more logical format:
Huxley 616 server room: as we've learnt from experience, promises to maintain a reliable power supply for the server room cooling systems have not always been kept and we have had a number of unscheduled cooling power outages over the period this facility has been in operation. Quite a lot has been done behind the scenes to improve internal monitoring & control systems leading up to the scheduled Easter power shutdowns in the Huxley building and the partial cooling failure on Easter Sunday was handled more gracefully than during previous incidents. Work is ongoing to improve room infrastructure management even further.
Maths compute cluster: all nodes now run Ubuntu 14.04 LTS Server Linux and macomp00 (formerly a Red Hat dev server) has been re-introduced to keep macomp000 company. Both servers are set up identically for users to test code and run small jobs on. More details...
MySQL server: to meet a requirement from a research group for SQL database facilities accessible from the Maths compute cluster, a new general access MySQL server has been set up hosted on macomp00.
More details...
Two new queues, matlabpar4 and matlabpar6: - have been added to cater for the smaller Matlab jobs using the matlabpool facility using 4 and 6 cores respectively. Since 4 or 6 free CPU cores are more likely to become available on a node at the same time than eight, this means Matlab parallel jobs queued to these new queues will spend far less time in the queue waiting for resources to become available than is the case with the existing matlabpar queue, which requires 8 cores. So if you are a Matlabpool user, you'll find submitting two smaller jobs to matlabpar4 will stand a far greater chance of executing soon after submission than a single job on the matlabpar queue.
More details...
New git and CVS servers: a git-hub service with web-based access in addition to the traditional access methods has been introduced on the share2 server alongside the existing svn service and a cvs server has also been added.
Other news: half of the Stats linux cluster has migrated from Fedora Linux to Ubuntu 14.04 and a group within this section now has a 15 node Hadoop cluster as well as an 8-server iSCSI SAN for working with large data sets.