GPG Cluster Usage
The cluster is integrated with the Unix network at Glasgow SoCS, so
you can log in to nodes and launch programs freely, e.g. ssh
gpgnode-01. External users who collaborate with the GPG are
welcome to use it. See details below.
Cluster Spec
This is primarily a 20 node (320 core) distributed compute server,
e.g. not a data server. There are no reliability capabilities, so no
Raid array or UPS.
Hardware
1. 20 nodes
2. 16 Intel cores/node (2 * Intel Xeon E5-2640 2GHz)
3. 64Gb RAM per node, i.e. 4Gb RAM/core.
4. Local Disk: 300Gb/node
5. 10Gb Ethernet Interconnect
Software
The operating system is Ubuntu 22.
Free-for-all Nodes
The free-for-all nodes are gpgnode-01, gpgnode-02, and
gpgnode-03. These are available for development and small scale
measurements with no prior booking.
External users can simply email Phil.Trinder@glasgow.ac.uk to obtain a
Glasgow login that will give access to these nodes.
Bookable Nodes
The remaining 17 nodes, gpgnode-04 .. gpgnode-20 are can be booked for
sole use as a single cluster. Why 17 nodes? Well experience also shows
that 1 node is often unavailable for some hardware/software/user
reason, and with 17 nodes we can reliably measure up to 16 nodes /256
cores.
Access to the remaining 17 nodes is
* restricted to members of a Unix group
* booked using a web-based calendar
* entails joining a mailing list where announcements are made.
Email Douglas Macfarlane and myself {Douglas.MacFarlane,Phil.Trinder}@glasgow.ac.uk if you would like access to these.
Booking Protocol
When you want to use the cluster would you
* Book free slot(s) in the Google calendar
* When your slot arrives have fun!
When booking slots please:
* Book only as many slots as you need, and for as long as you need
* Give your name
* Indicate either All nodes, or the subset of nodes you will use
Let Phil Trinder know of any issues (address above).
Additional Nodes
There are two additional nodes (gpgnode-21 and gpgnode-22), each equipped with 32 cores
(2 * Intel Xeon Gold 6426Y 2.5GHz), 128 Gb RAM, 1.2 TB storage on two local disks, and
10Gb Ethernet Interconnect.
These are bookable nodes as well, following the booking protocol outlined above.