Why aren't operating systems getting faster as fast as hardware?
Ousterhout 89: Why aren’t operating systems getting faster as fast as hardware?
Citeseer link: http://citeseer.ist.psu.edu/ousterhout89why.html
Reference count: Citeseer 175, Google Scholar 266.
Google Scholar does not have a hotlink to the paper (needs to be fixed).
Analysis.
What makes this a well-cited paper? Some attributes that I could see from a first and second read-through:
The systems being benchmarked spanned a number of years of systems design, and varied in raw performance by a factor of more than 10. All in all 6 different pieces of hardware with 4 different operating systems (giving a total of 9 systems) were tested, and the benchmarks run spanned the gamut from microbenchmarks of individual system calls all the way up to one large-scale benchmark. Hidden inside one relatively brief paper is a whole lot of data.
The question posed is an important one, and even now 16 years later is an interesting one: how well is software keeping up with hardware change? This is a perennial issue in computing systems design and requires careful experimental design and data analysis to expose.
This is the sort of paper that could be redesigned for SECSE07 and resubmitted using a modern (1999-2007) era range of hardware platforms and a similar design of micro to macro benchmark scale. The number of systems to choose from is larger now, and there are many more confounding variables, but this is still a compelling approach.
Technorati Tags: benchmarking, usenix, secse, performance
Comments