Hello! My name is Perry, hailing from Glasgow, Scotland. I’ve just completed my first degree, an MInf in Informatics, at The University of Edinburgh. For my summer of HPC, I will be working at the BSC, Barcelona.
I’m curious about the roles that computers can play in aiding people to take more direct control of their lives and communities, and am wary of the risks which they enable. My research interests are in machine learning and systems programming, and I am a dilettante of computer security.
In my spare time, I enjoy cooking, and exploring the cuisines of different places. I recently perfected my Hollandaise sauce, and I look forward to seeing what flavours my host city will inspire in me. I also enjoy many different types of music and movies, both old and new, and am constantly reading a wide range of books, both fiction and non fiction. (For years I’ve been maintaining lists of music, films and books to check out. The lists continue grow at a rate faster than I can consume.)
My graduation ceremony at Edinburgh University (woo!), so I arrived a day late to the training week in Bologna, and a delayed flight meant that in my first day I was a sleep-deprived gremlin. A good night’s sleep, and tactical consumption of coffee has seen me back to my usual self.
I’m part of a cohort of 25, and we are collected together in the city of Bologna, Italy, for a week of training. The previous cohort had their training week in Edinburgh, and though I like the city, I am grateful for the change of scenery.
This week, we are based in the CINECA, a top research supercomputing centre in Italy. Here, we have the opportunity to use one of the clusters, and familiarise ourselves with some of the HPC tool-chains. So far we have built examples which try to accelerate simple applications such as numerical approximations and manipulations of data objects too large to fit on a single node. Thus far, we have used MPI, and OpenMP, two parallel programming paradigms, with CUDA, and HPC data visualisation upcoming.
To perform the computations, we have been submitting jobs using the Slurm system. This seems to be a standard, as I used it on the GPU teaching cluster at The University of Edinburgh, a collection of 200 GPUs used by students taking courses in deep learning.
We were given a tour of the place. These machines consume a lot of power, and nearly half of that goes towards keeping them from overheating. Running them in a hot country only adds to that power bill…
Our guide gave us a brief technical overview of the black obelisks in the room ahead of us. Once we entered, all of their words were lost to us. An unending roar, regular enough to be almost relaxing, though near the point of being painfully loud. You can see a short video clip of the experience here.
At the end of the week, I’ll be heading to Barcelona, where I’ll be working at the Supercomputing centre there – BSC.
I’ve often heard that nowadays “computation is free”, and that the biggest overheads in modern workloads come from the costs of moving and accessing data. Consequently I will be researching the effects of different memory technologies on workload run time. For example, we might have a type of memory which has much faster read times, but very slow write times. If we leverage this, and place data which we use frequently, and change rarely, we should see improvements.
Dimitrios Voulgaris, a fellow student will be joining me in Barcelona. I have met Dimitrios, and look forward to getting to know him better in the coming weeks.
I will say more about my project as I learn more about it, and hope to present my findings in an interesting way. During the placement, I seek to gain more experience in profiling applications, and using the Valgrind toolsuite, which I believe researchers at my host centre have made valuable extensions to. I am looking forward to meeting the variety of talented individuals that live and work in Barcelona.
Thank you to PRACE for organising the program, and to my peers for being a welcoming multi-disciplinary group. Best of luck to you all!