A User-Level Remote Paging System for Grid Computing

Abstract

Most of grid applications require a large amount of memory resources. If there is not enough memory supplied during execution, the program performance will be significantly degraded by page swapping. To address this problem, we propose a remote paging system called Grid Remote Pager (GRP) for grid computing in this paper. Compared with other remote paging systems, GRP keeps high portability and transparency. The more important matter is that it actively adapts the limit of the amount of memory available for user programs according to the memory load of owner jobs. Consequently, not only the competition of memory between owner programs and user programs is effectively alleviated but also sufficient exploitation of free memory is achieved by user programs with the support of GRP.

An Adaptively Hierarchical Framework for Remote Paging on Grids

“…This system has a good scalability and allows page migration between memory servers to achieve load balance and reliability. In contrast, Wang [3] aimed at alleviating the negative impact from memory competition between resource consumers and providers on the performance of user programs by adaptive remote paging. However, these works did not put attention on the framework of memory servers in remote paging either.…”

“…Since the memory demand of data-intensive problems usually are much higher than the memory capability of any single machine even they are parallel processed, the time cost of resolving these problems are hugely increased due to an enormous number of disk I/O operations caused by page swapping. To address this problem, several systems such as RAM Grid [2] and Grid Remote Pager (GRP) [3] recently have been proposed to integrate the memory resources in grids for remote paging. With the support of these systems, the virtual memory manager can swap pages between local memory and remote memory instead of local hard disks.…”

An Adaptive Remote Paging System on Computational Grids

“…Different to RAM Grid, Grid Remote Pager [5] adopts a centralized framework for the client-to-server matching, and progressively adapts the memory utilization of user programs to reach a balance point between avoidance of memory competition and exploitation of free memory while it adapts the memory utilization of only the memory clients but the memory servers.…”

“…The dynamic memory allocation and release and the page fault handle in AGRP have been clearly described in the paper [5], the rest of this section will be focused on describing the allocation policy and the selection policy of AGRP, as well as the adaptation in the memory utilization of the memory servers.…”

“…Rui Chu [4] developed a peer-to-peer based RAM grid to achieve good scalability and reliability in remote paging. Po-Sen Wang [5] developed a grid remote pager which can dynamically adapt the maximal amount of physical memory allowed for user programs to avoid memory competition and minimize the number of page swapping. However, several important issues specific for remote paging in grids did not receive attention from these works.…”