Telecomms research sped up with GPUs
Cognitive radio researchers at Trinity College, Dublin have increased the speed of their simulations using multicore computation and GPUs. The research team at the CTVR telecommunications research centre, based at Trinity College Dublin, are developing algorithms for building future phone networks that can organise themselves.
Traditionally, cellular phone companies like Vodafone, O2 and T-mobile bid against each other for exclusive access to radio 'spectrum'. It is a slow and expensive process that often ends up with spectrum being under used. The research team at CTVR are exploring ways to avoid the need to carve up spectrum in advance.
To test their algorithms, the researchers needed to build a simulation model to show they would work in practice. 'To test our algorithms, we wanted to simulate a reasonably-sized network and try out many different combinations of transmitters and receivers to be sure that equilibrium is reached in all cases,' said researcher Dr Irene Macaluso. 'A "reasonably sized network" means individual testing of many hundreds of thousands of cases and our typical analyses could take several days using sequential programs.'
The CTVR team used Xcelerit's software development kit to split the tasks over the multiple processor cores. Xcelerit provides software tools to make many-core processors accessible to mainstream programmers.
On a PEER 1 system with two Intel Xeon E5620 CPUs it was possible to speed up the code by a factor of 13 compared to a sequential implementation using a single CPU core. When two Nvidia Tesla M2050 GPUs were added, speedups of 140x were recorded without further code changes.
'This meant that some of our simpler computations were completing in under a second and our typical simulations went from 35 hours down to just 15 minutes,' said Dr Macaluso. 'Sourcing the extra compute horse-power from the cloud was great because we had no lead time for new hardware and we can rent it whenever we need it in future.'
Xcelerit’s CEO, Hicham Lahlou, commented: 'The CTVR team was able to make minor changes to their simulator implementation to make it work with our Xcelerit SDK. Once that was done, they could run it on different machine configurations including GPU hardware from Nvidia.'
Xcelerit is planning to deploy its toolkit widely in education and research. 'Universities and research centres are great customers for us,' said Lahlou. 'They really stretch our product's capabilities and are very influential when it comes to industry take-up.'