Advance warning of scheduled systems work and downtime

November 27th: Bazooka Hadoop cluster planned shutdown for network switch replacement & minor server upgrade

The Bazooka cluster will be shut down on Friday November 27th for the network switch on the gigabit internal network to be replaced. A few ports on the existing 24-port switch are faulty with one not working at all and some others only negotiating a 10 Mbit/s half duplex link. This does not prevent the cluster from functioning since this network is only used for internal control traffic and DNS looksups, etc which re-route to the 10 gigabit network in the event of problems but mixing control and Hadoop-related traffic on the high speed network does impact performance to some extent. The opportunity will be taken to fit a larger and more recent 48-port unit with a faster switching backplane.

At the same time, the bazooka server will be upgraded with the full iDRAC Enterprise hardware, replacing the existing basic iDRAC on-board remote management facility. For some reason the latter does not work via the Huxley 616 management LAN so bazooka cannot be remotely controlled through this route, meaning it can't be powered on remotely from a completely-off cold state, for example.

The cluster is expected to be down for 1-2 hours and interested users have already been notified.

No other work is scheduled for the forseeable future

Andy Thomas

Research Computing Manager,
Department of Mathematics

last updated: 26.11.2020