Global Memory Management for a Multi-Computer System
In this paper, we discuss the design and implementation of fault-aware Global Memory Management (GMM) for a multi-kernel architecture. Scalability of today's systems is limited by SMP hardware, as well as by the underlying commodity operating systems (OS), such as Microsoft Windows or Linux. High availability is limited by insufficiently robust software and by hardware failures. Improving scalability and high availability are the main motivations for a multikernel architecture, and GMM plays a key role in achieving this. In our design, we extend the underlying OS with GMM supported by a set of software failure recovery modules in the form of device drivers. While the underlying OS manages the virtual address space and the local physical address space, the GMM module manages the global physical address space. We describe the GMM design, prototype implementation, and the use of GMM.
author = {Dejan Milojicic and Steve Hoyle and Alan Messer and Albert Munoz and Lance Russell and Tom Wylegala and Vivekanand Vellanki and Stephen Childs},
title = {Global Memory Management for a {Multi-Computer} System},
booktitle = {4th USENIX Windows Systems Symposium (4th USENIX Windows Systems Symposium)},
year = {2000},
address = {Seattle, WA},
url = {https://www.usenix.org/conference/4th-usenix-windows-systems-symposium/global-memory-management-multi-computer-system},
publisher = {USENIX Association},
month = aug
}