Scope consistency: a bridge between release consistency and entry consistency
Proceedings of the eighth annual ACM symposium on Parallel algorithms and architectures
Optimizing Home-Based Software DSM Protocols
Cluster Computing
The Affinity Entry Consistency Protocol
ICPP '97 Proceedings of the international Conference on Parallel Processing
Adaptive Prefetching Technique for Shared Virtual Memory
CCGRID '01 Proceedings of the 1st International Symposium on Cluster Computing and the Grid
An Efficient Lock Protocol for Home-Based Lazy Release Consistency
CCGRID '01 Proceedings of the 1st International Symposium on Cluster Computing and the Grid
Efficiently Adapting to Sharing Patterns in Software DSMs
HPCA '98 Proceedings of the 4th International Symposium on High-Performance Computer Architecture
A practical transparent data sharing service for the grid
CCGRID '05 Proceedings of the Fifth IEEE International Symposium on Cluster Computing and the Grid (CCGrid'05) - Volume 2 - Volume 02
| Hi-index | 0.00 |
In this paper we examine the approach of enabling parallel processing of general applications on top of Grids by adapting and extending methods that have been used successfully for tightly coupled systems. Our approach is based on the fundamental design of the Software Distributed Shared Memory (SDSM) model which we use in order to achieve the performance objectives of our environment. In this work we propose a new dynamically Reduced Message Protocol (RMP) for DSM systems, which adopts the scope-consistency model[8]. First, we outline our basic RMP protocol and then we propose two enhancements: the RMP-MH (Multiple Home)and the RMP-LM (Lock Migration). Our experimentation with RMP resulted in significant improvements by reducing the total messages and the overall number of page faults. In order to demonstrate the efficiency and the effectiveness of the RMP protocol we used SPLASH as well as synthetic application benchmarks, to highlight the protocols advantages, demonstrating a reduction in the total execution time of up to 30% for SPLASH applications and up to 61% for the synthetic application benchmarks.