We have closed account creation some time ago when we reached our targeted performance. We currently have 12 simulation runs going that estimate model parameters for different model formulations. Fitting the parameters takes a bit longer for any particular model than we had anticipated. However, we cannot speed up this process by simply increasing the number of host by allowing new users to sign up. The reason is that it is hard to find a good optimization algorithm in an environment which is highly parallel and where the turnaround time for workunits is relatively long and highly variable. This means that the gain from computing more workunits in parallel is not very great at the moment as we have hit the limit of parallelization of what we are trying to do. Therefore we will only open for more users once:
-We can further improve the optimization algorithm, or
-We have a more simulation runs to run simultaneously, or
-Performance drops because of user attrition.
In between the runs that fit parameters to data we run batches of workunits that make predictions of control interventions: We currently study the impact of different vaccine types on the epidemiology and their cost-effectiveness if distributed via a number of different distribution modes. We explore the effect of using different model version on these predictions as well as the effect of making different assumptions about some of the vaccine parameters (their protective efficacy, the duration of the protective effect, the population covered by the intervention etc.) malariacontrol.net thereby only generates the prediction of the impact these interventions have on the epidemiology. Here in Basel we have a database application (written in Ruby using the Rails framework) which links these results to cost data to predict the cost effectiveness of these intervention strategies. This part is computationally cheap compared to making the epidemiological predictions. We are currently analyzing the first completed set of predictive runs.
We are now at a stage where we can consider adding support for a more platforms. We plan to port the science application to Mac and Linux 64-bit. We have ordered the necessary hardware and compilers and will start to work on this within the next few weeks.
The map predictor application, which we have restarted a few days ago, seems more stable but still doesn’t run well on all hosts. The reason for that is not the science application itself, but the BOINC wrapper, which sometimes fails to copy the input file to the slot directory. I’ll post some information on how we want to proceed with this shortly.
Swiss Tropical and Public Health Institute