The article presents a reflective examination of what has worked in computer system research, as opined by Butler Lampson in 1999 and reevaluated in the modern context. Noteworthy is the persistence of previous ‘Yes’ categories and the evolution seen within ‘Maybe’ and ‘No’ classifications, with particular attention on developments in parallelism, RISC, garbage collection, and capabilities. The conclusion underscores the volatile nature of predicting future technological successes and recognizes the impact of computer systems research over nearly two decades.
Main Points
Enduring relevance of foundational concepts and shifts in adoption
The enduring relevance of concepts labeled ‘Yes’ in 1999 showcases the foundational aspects of computer systems research, while the movements observed in the ‘Maybe’ and ‘No’ categories signal shifts in technology and adoption over time.
Insights
Parallelism’s status as a 'Maybe' indicates the complexities and challenges that have persisted in fully leveraging it despite technological advancements.
This is, unfortunately, still a Maybe. Between the end of Dennard scaling and the continued demand for compute, chips now expose plenty of the parallelism to the programmer. Concurrency has gotten much easier to deal with, but really extracting anything close to the full performance available isn’t much easier than it was in 1999.
RISC was initially expected to dominate but ended up being overtaken by the prevalence of x86, illustrating shifts in the computing architecture landscape.
If this was a Maybe in 1999 it’s certainly a No now. In the 80s and 90s a lot of folks, probably the majority of folks, believed RISC was going to take over the world and x86 was doomed.
Garbage collection is more widely adopted now, highlighting its importance in modern programming languages for memory management.
This is a huge Yes now. Every language that’s become successful since 1999 has GC and is designed for all normal users to use it to manage all memory.
The shift from viewing capabilities as a 'No' to a 'Yes' reflects the evolution and acceptance of security measures in mobile operating systems.
This post originally had Capabilities as a No in 2015. In retrospect, I think that was a mistake and it should have been a Yes due to use on mobile.
Links
- past and future of “computer systems research”
- end of Dennard scaling
- Erik Meijer and Butler Lampson talked about this
- recently added features that make it easier to take advantage of trivial parallelism
- explanation of the Alpha design
- James Morris wrote
- driver verification tool
- very little impact on industry
- things are changing faster now than they were in 1999