Evaluating Task Migration Overhead for Heterogeneous Multi-Core Architectures
Date Issued
2008
Date
2008
Author(s)
Chen, Kuan-Ju
Abstract
A heterogeneous multi-core architecture is composed of cores of varying size and performance. Many studies show that single-ISA heterogeneous multi-core architectures can provide higher performance than homogeneous multi-core architectures. Because a program may need different resources during its execution, it can use system resources more efficiently by moving to a core that matches its resource demands in a heterogeneous multi-core architecture to improve the overall performance. We design a program scheduling policy to determine the assignment of programs to the cores for improving the overall performance by maximizing the instruction consumption for every execution period for each core.owever, when a program migrates between cores, some of the data in the original private cache must also move to the new core, which causes extra overhead. To solve this problem, we propose a new cache architecture called SwitchL2, which can be configured to operate in a private or a shared mode. In the private mode, a core may only access its own L2 cache, but it can access L2 cache of other cores through the switch. In the shared mode, the core can access all L2 caches simultaneously. Which mode to use depends on a program’s usage of L2 cache. Compared with traditional L2 cache architectures, SwitchL2 offers 1% performance improvement and reduces 50% dynamic energy consumption and 3% total energy consumption of L2 caches for the SPEC benchmark programs in our simulated study.
Subjects
homogeneous multi-core
heterogeneous multi-core
L2 cache architecture
performance
energy consumption
SDGs
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-97-R95922010-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):f3a2640e70c9b71edd938942c51a23f0