POCL-R: Distributed OpenCL Runtime for Low Latency Remote Offloading

Running complex applications on mobile devices stress their performance and power consumption limits. As new wireless networking technologies that promise lower communication latencies appear, it is interesting to study whether dynamic offloading from mobile devices across a network to a nearby compute cluster is feasible when quality and complexity can be traded off dynamically. This type of scheme calls for a light weight heterogeneous distributed programming layer.

Here we describe our work-in-progress distributed OpenCL runtime optimized for low latency quality-complexity tradeoff cases. We call it POCL-R1, since it is implemented on top of the Portable Computing Language (POCL [2]) as a device layer implementation that exposes remote compute devices accessible over a network connection in a transparent manner on the local OpenCL platform. In order to improve the latency, we expand upon our earlier work on exploiting OpenCL-described heterogeneous task parallelism [3] for intelligent cross-device event synchronization and buffer management.

The closest previous works to ours are SNUCL-D [4], which duplicates the entire execution flow of the host program on the remote computer, and rCUDA [1], which, being based on the CUDA API, has a more limited hardware support. All the previous projects we found are mainly focused on throughput rather than latency, and also appear to be inactive at the time of this writing. In contrast to coarse grained frame/game streaming solutions, our solution allows fine grained partial offloading of the application at OpenCL kernel level. The aim of POCL-R is to make it easy to switch between local and remote compute devices on the fly, which is beneficial when the mobile device suffers from a connection loss, or roams between between networks with different available compute resources (see Figure 1).