Farewell for now
By val on Sep 21, 2004
I learned a few things over the past few months. Sun is a great working environment; most other companies score much higher on the Dilbert scale. Which explains why I know so many people who left Sun and then came back a few years later (an option I will consider). After deciding I wanted to quit, I stopped doing all the things I didn't want to do - and ended up being more productive. Finally, a little patience (and negotiating skill) will go a long way (unfortunately, these are two qualities I am notoriously short on).
I feel lucky to have had the chance to work on ZFS. It really is a groundbreaking new file system - and so easy to use. Now I just hope that it gets ported to Linux soon, so I can use it.
I may start a new weblog; if so, it will be linked to from my home page:
After leafing through the stack of computer science papers on my desk, looking for one last systems paper to talk about, I find I can't pick just one. Here for your reading enjoyment, then, are four of my favorite systems papers.
Aside from the awful title, this is a fantastic paper. The authors managed to automate (in a practical, easy-to-understand manner) checking for common, simple bugs such as forgetting to drop a lock on error exit from a function. They found many bugs in widely-used real world systems with relatively little human effort. I've seen their stuff in action; it's good. And as usual for an Engler paper, it's well-written and a pleasure to read.
Don't you hate it when your laptop's screensaver starts up in the middle of a presentation? Angela Dalton and Carla Ellis explore using low-power sensors to make power-saving smarter. Angela had a great demo at HotOS: a laptop with attached camera that turned off the display when no human face was visible. All day long, I saw people sitting in front of a laptop and then suddenly ducking to the side, again and again, testing her demo. I like that they are taking a creative new approach to power-saving, instead of fiddling with timeouts and mathematical models.
Lots of research has been done with the idea that users will annotate files with performance hints using some obscure system-specific interface (yeah, right!). In this paper, Daniel Ellard shows that programs are already giving the system useful performance hints - by the names they give to files. Even more exciting, filename patterns and usage patterns can be automatically correlated by a modeler program.
I don't really need to describe this paper, do I? I just wanted to explain why I like this paper so much. Many people have tried write a distributed file system that is correct, generic, and performs well in all cases - and are still trying. What I like about GFS is that they picked a very specific problem (large scale distributed processing using a queuing system) and solved it in a specific way.
I hope you enjoyed the Confessions; thanks for reading!