Review - A Case for NOW

Commodity, off-the-self computing components offer a higher cost-performance
over custom designed systems (MPPs, mainframes, supercomputers). The
challenge is to build large scale computing systems out of mass-produced
components in order to achieve comparable performance at a fraction of the
cost.

Anderson et al. do a good job in arguing that, at the time of publication,
commodity computing components (RAM, bandwidth, etc.) had reached a point
where they could be used to develop large scale computing systems. The paper
presents a fair cost analysis highlighting the requirements for performance
comparable to that of a MPP. Their research concludes that a scalable
network, a parallel file system and low communication overhead are the
primary consideration for high performance in a NOW system. Inherent
challenges in the model are also discussed, where I/O bottlenecks seem to
dominate.

The primary weakness of this paper is its assertion that the operating
system must be altered at the kernel level for NOWs to be successful. This
is in direct conflict with the premise of the paper, which is to build MPP
comparable performance on mass-produced components. Although it is obvious
that integration software is required, the argument is weakened by this
section of the paper.

The incorporation of sociological considerations into this research paper is
well thought out and thorough. I found that this section contributes a great
deal to the use of the NOW system in practice.