(no subject)

Brian TAM (btam@cs.ucsd.edu)
Tue, 30 May 2000 05:26:04 -0700 (PDT)

"Implementing Global Memory Management in a
Workstation Cluster"

Operating systems on workstation clusters typically
operated as a collection of loosely-connected processors,
each one operating autonomously and independently of one
another. This paper describes a global memory management
system that aims to exploit cluster-wide memory to
improve the performance of memory-intense applications.
Called the Global Memory Service, this facility swaps
pages between local and remote nodes on faults, electing
candidates by age. An implementation is described in
which most features of the system were incorporated into
an existing OS, the OSF/1, and performance measures were
taken using memory-intensive benchmarks. The authors
assessed their system favorably by demonstrating speedup
in performance with increased cluster-wide memory and
increased performance despite heavy network traffic or
sharing of pages on common nodes.

I find it odd that the authors did not report
application run times for GMS. Speedup measures are
less meaningful without baseline values to compare with.