[CSC2231] Paper Review: A Case for NOW

This paper suggests the use of networks of workstations (NOW) to replace
massively parallel processors (MPPs) for various computing tasks. NOW
offers better cost-performance ratio than a comparable MPP system with
similar resources (in terms of memory, storage and computing power). The
authors implement the Berkeley NOW project to demonstrate the
feasibility to use NOW.

Section 3 of the paper discusses the potentials to use the idling
memory, storage, and computing resources cooperatively in a NOW
environment. These ideas are backed up with the Berkeley NOW project
discussed in section 4: GLUnix provides networked memory and parallel
computing infrastructure and xFS provides a distributed RAID storage
system. I find this platform valuable because non-cluster-aware
applications can start exploiting the power of a cluster of machines
transparently.

The main weakness of the paper is the lack of practical evaluation of
the performance of NOW. This leads to the issue regarding the
difficulties to port existing applications to the Berkeley NOW platform
in practice. The authors might have run the GATOR simulation on their
NOW platform if it is obvious to port application; however, this might
not be the case. In addition, this paper does not discuss the ongoing
cost to house a NOW platform. For example, it is reasonable to compare
the cost of maintenance and the cost of power consumed by a NOW
environment versus a MPP system.
Received on Thu Sep 15 2005 - 06:30:32 EDT