Randomized algorithms
Proceedings of the fourteenth annual ACM symposium on Principles of distributed computing
Optimistic parallelism requires abstractions
Proceedings of the 2007 ACM SIGPLAN conference on Programming language design and implementation
Collaborative threads: exposing and leveraging dynamic thread state for efficient computation
HotPar'10 Proceedings of the 2nd USENIX conference on Hot topics in parallelism
Proceedings of the FSE/SDP workshop on Future of software engineering research
Efficiently speeding up sequential computation through the n-way programming model
Proceedings of the 2011 ACM international conference on Object oriented programming systems languages and applications
Exploiting coarse-grain speculative parallelism
Proceedings of the 2011 ACM international conference on Object oriented programming systems languages and applications
Multi-solver support in symbolic execution
CAV'13 Proceedings of the 25th international conference on Computer Aided Verification
| Hi-index | 0.00 |
With the advent of multi-cores and many-cores, traditional techniques that seek only to improve FLOPS of performance or the degree of parallelism have hit a roadblock with regards to providing even greater performance. In order to surmount this roadblock, techniques should more directly address the underlying design objectives of an application. Specific implementations and algorithmic choices in applications are intended to achieve the underlying realism objectives in the programmer's mind. We identify two specific aspects of this realism that traditional programming and parallelization approaches do not capture and exploit to utilize the growing number of cores. The first aspect is that the goal of minimizing program execution time can be satisfactorily met if the program execution time is low with sufficiently high probability. We exploit the fact that randomized algorithms are available for many commonly used kernels, and that the use of parallelism can achieve very low expected execution times with high probability for these algorithms. This can provide speedups to parts of the application that were hitherto deemed sequential and ignored for extracting performance via multi-cores. The second aspect of realism that we exploit is that important classes of emerging applications, like gaming and interactive visualization, have user-interactivity and responsiveness requirements that are as important as raw performance. Their design goal is to maximize the functionality expressed, while maintaining a high and smooth frame-rate. Therefore, the primary objective for these applications is not to run a fixed computation as fast as possible, but rather to scale the application semantics up or down depending on the resources available. Our framework intends to capture the responsiveness requirements of these applications as they pertain to expressed realism and automatically scale the application semantics expressed on every architecture, including very resource-rich many-cores.