| सारांश: | Mobile Cloud Computing (MCC) facilitates energy efficient operations of mobile devices
through computational offload. The mobile devices offload computations to nearby cloud
servers while limiting energy consumption in the low-power wait mode. The MCC offloading
frameworks are enabled by system virtualization, application virtualization, and
native code migration techniques to address the heterogeneous computing architectures.
The existing MCC offloading techniques suffer from either computational or communicational
overheads leading to higher execution time and energy consumption on the
cloud server. This research work addresses the overhead of conventional MCC offloading
frameworks while focusing on vectorized applications based on Single Instruction
Multiple Data (SIMD). We propose SIMDOM, a framework for SIMD instruction translation
and offloading in heterogeneous MCC architectures. The SIMD translator utilizes
re-compilation of SIMD instructions of the mobile device (ARM architecture) that are
translated to corresponding cloud server instructions (x86 architecture). Based on inputs
from the application, network, and mobile device energy profilers, the offloader module
decides upon the feasibility of code offload. The SIMD translator is analyzed for its accuracy
and translation overhead. The impact of code offload size, application partition, and
device sleep time is investigated on the energy and time efficiency of the mobile applications.
The lower feasibility bounds for server speed and application partition are derived
from the system model. The SIMDOM framework prototype is implemented on a cloudlet
and a cloud server. Results show that SIMDOM framework provides 85.66% energy and
3.93% time efficiency compared to MCC-disabled execution. Comparison with state-ofthe-
art code offloading framework reveals that SIMDOM provides 55.99% energy and
57.30% time efficiency. The SIMDOM framework provides 31.10% higher energy efficiency
while translating SIMD instructions as compared to existing MCC offloading frameworks. The improvement in energy and time efficiency increases the usability of
MCC offloading frameworks for vectorized applications.
|
|---|
