[Revised & updated, Tuesday 19th January 2021]
The information in this blog post is relevant to anybody who manages or uses a self-managed server located in the Informatics Forum self-managed server room IF-B.Z14.
It does not relate to equipment located in the smaller Forum self-managed server room IF-B.01 – so anybody who manages or uses equipment in that room only can safely ignore the rest of this post.
Currently, all electrical circuits in IF-B.Z14 are rated at 20A – which is a limitation given that the power bars we use in the room are rated at 32A. We have recently had eight new 32A circuits installed in the room, and we would now like to swap the input power feeds of one of the two power bars fitted to each rack from their existing 20A circuits, to new 32A circuits.
In order to preempt the likelihood of 20A circuit breakers being blown (as has already happened over the past year, on account of usage patterns in the room) we would very much like to do this work before attempting to install any further equipment in the room.
We intend to swap the the power feeds of the left-side PDUs (looking from the rear of the racks) in seven of the eight racks. (We’re currently excluding Rack 0 from the work, since we’ve been specifically asked to avoid a power loss to one of the machines in that rack for now.)
Arranging the swaps will necessarily mean a short loss of power to each of the power bars involved. As a knock-on effect, there will be a short loss of power to any server which is powered via a single power supply unit (PSU) fed from the power bar in question. Servers equipped with multiple PSUs shouldn’t be affected.
We currently propose doing this work in the afternoon of Thursday 21st January, starting at 1pm. Would all managers and users of self-managed servers located in the Informatics Forum self-managed server room IF-B.Z14 please consider the above proposal, and contact me if this date and time will cause them a particular problem? Thanks.
To emphasise: servers which are affected will completely lose power for a short period – so it would be best if they were not in active use at the time, and best of all if they had been shutdown in advance.
Within limits, we will be able to shutdown servers ourselves just ahead of the work, and restart them again afterwards, provided that the servers involved have been configured by their owners to perform an orderly shutdown when their physical power buttons are pressed.
What follows is a list of servers in each rack which will be affected:
bob chilli dimsum imola monza tofu turingfpga
delrubio ideal4 vali
nsl000 nsl002 nsl003 nsl005 nsl007 nsl009
austin chicago indianapolis ottawa sanantonio toronto Mellanox-SX6012