TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

Accelerating Computation on an Android Phone with OpenCL Parallelism and Optimizing Workload Distribution between a Phone and a Cloud Service

Tutkimustuotosvertaisarvioitu

Yksityiskohdat

AlkuperäiskieliEnglanti
Otsikko2016 Intl IEEE Conferences Ubiquitous Intelligence & Computing, Advanced and Trusted Computing, Scalable Computing and Communications, Cloud and Big Data Computing, Internet of People, and Smart World Congress (UIC/ATC/ScalCom/CBDCom/IoP/SmartWorld),
KustantajaIEEE
Sivut636-642
ISBN (elektroninen)978-1-5090-2771-2
DOI - pysyväislinkit
TilaJulkaistu - 18 heinäkuuta 2016
OKM-julkaisutyyppiA4 Artikkeli konferenssijulkaisussa
TapahtumaIEEE International Conference on Scalable Computing and Communications. -
Kesto: 1 tammikuuta 2000 → …

Conference

ConferenceIEEE International Conference on Scalable Computing and Communications.
Ajanjakso1/01/00 → …

Tiivistelmä

We evaluate workload distribution optimization between an Android phone, a cloud service by considering the overall impact of both computation, data transfer. We use OpenCL parallelism on Android to obtain high computation performance. We implement an escape time algorithm to compute the Mandelbrot set with OpenCL,, with Java as a reference for comparison. In an experiment of setting the escape boundary at 256, OpenCL offers about 5.0X to 7.5X faster computation compared to Java. With a cloud service, data transfer becomes a dominant factor when the amount of computation is low. In a set of four experiments of sharing workload of computing the Mandelbrot set between a cloud service, a phone, data transfer consumes on average over 80% of processing time. In those experiments computing locally with OpenCL on an Android phone yields faster processing time. On the other hand, local computation capacity becomes a bottleneck when the amount of computation is high. With the escape boundary at 65536, requesting computation from a cloud service yields up to 7.55X speedup.