A company called Descartes Labs has become number 136 on the top500
supercomputer list just by paying for a large enough collection of nodes
on its Amazon Web Services account
However, Dr Paul Calleja, directory of the University of Cambridge
Research Computing Services, argues that this is not really
cost-effective compared to buying your own actual gear, and the I/O
performance is not that good anyway. Such a cloud cluster may score
well on LINPACK, but not so well on real-world supercomputing problems.
Dr Calleja advocates using OpenStack to build a “hybrid” combination of
on-premise physical hardware backed up by cloud services for those
times when you need a bit of a (temporary) boost.
And for learner supercomputer drivers, there is always the option of a
Raspberry π cluster to practise on, before being allowed behind the
wheel of a real big-iron machine ...
Show replies by date