A Case for Redundant Arrays of Inexpensive Disk (RAID) (Patterson, 1988)

Jonathan Ledlie CS736, February 16, 2000

Patterson, et. al., provides an elegant solution 
to the "pending I/O crisis" in his idea for making the sum greater than its parts.  In 
fact, his paper is so admired that the database community's SIGMOD has given it its 
"Test of Time" award.  Beyond the actual implementations of the levels of RAID, this 
paper has at least three strong points.  First, it accurately recognizes the 
impending bottleneck that users will soon face: as the processor and RAM are 
getting faster and cheaper, disks, particularly seek times, are remaining nearly 
constant.  The ramification of this is that even though we believe that our 
applications could be running hundreds of times faster, they can only run one order 
of magnitude faster, due to the I/O bottleneck.  I found this in interesting 
juxtaposition to the Impact paper which argued that RAM and cache performance is a 
significant bottleneck.  Second, instead of focussing solely on performance, as 
most operating systems evaluations do, Patterson's idea takes cost into account. 
Cost, unlike ease-of-use, is somewhat easily quantifiable, but is more often than 
not, left out of the equation.  Third, in evaluating their idea, the Berkeley group, 
clearly differentiates between the two main kinds of troublesome I/O loads: huge, 
supercomputer files, and lots of random non-repeated seeks, like in database 
applications.  They leave the reader with a relatively clear feeling as to what to 
choose for her particular installation.  One problem with RAID is that, even though 
the cost of the hardware for buying the drives will be less, it is a complicated 
system, and the cost of your RAID maintainer will be higher - the hardware is 
cheaper, but the human cost in expertise is probably more.