Members of Research IT’s Research Infrastructure Group have been working with the Proton Therapy research group (PRECISE) led by Prof. Karen Kirby at The Christie Hospital to improve clinical dose prescriptions in radiotherapy.
PRECISE Research Associate Nicholas Henthorn is investigating the biological effects of various types of radiation upon human cells. Whilst, it is known that different human cells exhibit different sensitivity to radiation, it is less known if there may be any geometrical mechanisms driving this.
To investigate this, a series of models based on the measured geometry of realistic human cell types are constructed using an in-house markov-chain polymer optimisation platform called G-NOME. These “cells” are then used with the open source Geant4 (GEometry ANd Tracking) code which can simulate the passage of radiation through each cell model.
Each cell model is irradiated to a prescribed dose and the DNA damage to that cell is scored. This is repeated 50 times per cell with a typical repeat taking 20 minutes. This means that each job takes approximately 17 hours running on a single CPU and Nick had 8,800 jobs to run!
Scoring the interactions between radiation and DNA allows for the prediction of DNA damage for a given radiation type. Ultimately, this model should be able to inform clinical dose prescriptions in radiotherapy allowing for more accurate and safer cancer treatments.
The 8,800 jobs were independent of each other and individually had modest computational requirements so Daniel Corbett, Research Infrastructure Engineer, recommended that the jobs should be run on the High Throughput HTCondor service offered by Research IT. As the simulations did not involve sensitive data they could also be burst into the Cloud from the Condor pool. This allowed the jobs to be run quickly allowing Nick to gather the required results for publication.
Having the ability to burst to the Cloud allows Research IT to dramatically and dynamically increase the capacity of the Condor pool while taking advantage of the cheap Spot prices offered by AWS. It is expected that Nick’s 8,800 jobs would use about 150k core hours, on the CSF this would take approximately 60 days. Taking advantage of Cloud bursting Nick was able to run all these jobs in 9 days – i.e. roughly 680 CPU cores running continuously.
If you are interested in using our High Throughput HTCondor service you can find further details on our website.