ma-backup4 was added today and is already mirroring data from one of the three compute servers used by the joint Maths:Physics research group. This joins the three existing on-site backup servers ma-backup1 to ma-backup3 inclusive which are now getting very full. Using the same architecture as the existing servers, ma-backup4 has 32 TB of installed disk to provide 24 TB of usable resilient storage. The offsite server in Slough, ma-offsite2, which also hosts ma-offsite3 will be expanded to to create ma-offsite4.
These four backup servers backup user data from all of the various compute servers - most of these have fast local disk storage facilities in addition to networked storage, which needs to be regularly backed up.
February 26th onwards: internal network expansion, phase 2
There are 110 college network connections in the Maths server room, Huxley 616, with over 170 systems installed and recent requests to host yet more systems that need to be accessible over the college network. Installing more college network sockets in the room is no longer an option owing to the disruption it will cause and also, the cost; there are no spare ports left on the existing network switches in the level 6 comms cupboard so at least one new switch plus a rack to contain it will be required which is expensive.
A dormant 'background project' that began about two years ago to move systems that did not really need college network connections over to dedicated internal networks and, at the time, freed up about 10 college network connections, has restarted in earnest. We are now moving more systems to internal networks which involves installing additional network switches, cabling and where required, extra network interfaces into systems that do not have sufficient network ports.
Work is ongoing but since February 25th, four college network connections have been released for use by the imminent ma-backup4 and new GPU servers and a proposed private storage server.
Coming soon: a new GPU server
A new GPU server has been delivered which will greatly improve our GPU computing facilities. Fitted with dual 16-core Xeon CPUs, 1.5 terabytes of memory and 14 hard disks providing a total usable local storage capacity of ~30 terabytes, eight nVidia RTX 2080i GPUs each containing 4352 GPU cores and 11 GB of memory are fitted. With support for nVidia's cuDNN (Deep Neural Network) libraries, this new addition is expected to become very popular.
We are now awaiting completion of electrical power feeds in Huxley 616, the installation of another 8 kVA UPS and the delivery of a rack before the new GPU server can be installed.