Mario Antonioletti, EPCC
EPCC, at the University of Edinbugh, commenced its outreach involvement with it’s attendance of the British Science Festival in 2012. Our aim was to promote and demistify what supercomputers are and how they are typically used to a general audience regardless of their age. To achieve this we chose an illustrative number of activities: we connected to the then UK national HPC service, HECToR, where we run a molecular dynamics application in parallel whose results could be visualised in real time and interacted with via VMD. This allowed the simulation to be externally controlled by participants who were then tasked with extracting a pheromone molecule embedded inside a mouse urinary protein which provided our tag line: “Can you make the mouse wee smell?”. We also invited participants to take on the role of processors to perform a number parallel sort of a set of integers printed on pieces of card randomly distributed amongst them using message passing methods – we talked them through the resulting algorithm. Finally we displayed a number of motherboards from previous actual supercomputers to show the main operational principles. At the time we were quite pleased with the results of our first event and a subsequent follow-up.
Over the following years we gradually came up with new activities and slowly evolved or replaced existing ones. For instance, logging on to a supercomputer from a science festival venue meant that you would need a reliable network to be available at a venue site which was not always the case. In addition, with no real baseline expectation of how fast an application should run, i.e. serial vs parallel, the resulting experience could be somewhat underwhelming. We thought it would be nice to take an actual supercomputer on tour with us but that would introduce insurmountable logistic problems nevermind the expense. If not a supercomputer then a model one might do but it was not until the follow-on UK national supercomputing service, ARCHER, came along that some funding became available for us to realise our dream – Wee Archie was born:
An 18 Raspberry-Pi based model of a supercomputer. Ok, it cannot compare to an actual modern supercomputer but all the essential elements are there with the bonus that you have flashing LED lights and you just cannot have a supercomputer without these! A number of applications have been developed to work in this environment: dinosaur racing, designing a plane’s wing to see whether you can generate enough lift to take off and a weather demo and we can now ship off our very own supercomputer to events. In fact demand for the use of Wee Archie within our proved high enough for a second system to be built this time using blue LEDs. Understandably not everyone will be able to build such a system but the experience of constructing the system allowed us to produce some instructions on how to build a Raspberry-Pi cluster.
Similarly the message passing parallel sort was not so amenable to the high throughput environment as presented at a science festival. The activity had to be instructor led and required several steps before the number list was globally sorted. We wanted something that would be more self explanatory and which would show the benefits of parallelism, the basis of supercomputers, with greater immediacy. We slowly iterated to what has become my personally favourite activity – coloured object sorting (I say object because we started with balls but we now use coloured bean bags). To do this we get one person to sort as many coloured objects as they can in 30 seconds by putting a coloured objects into the corresponding coloured bucket. We then get 2 or more people, a peer or a family group, to perform the same task including the single person sorter if possible – invariably they will get more and there lies the principle: you can do more within the same period when compared to one. We also get participants to record their score by putting on a sticky dot on a piece of graph paper, as shown below. This allows people to see how they performed in relation to others, sometimes engendering a competitive spirit, but it also allows you to draw many beautiful analogies, e.g. as the number of people increases you get contention so performance does not scale linearly, the variability in performance when the processors are not homogeneous (4 year old child is joined by mother), etc.
We have developed and tried other activities and have started writing these up. We are always looking for new activities or ways of improving existing ones. If you would like to collaborate with us to this end then please visit our GitHub pages. The overall aim is to inform and educate people about supercomputers and how they are used.