next up previous
Next: Rebutia Up: Sessions 7 & 8: Previous: Saguaro

Rhipsalis

The presenter for the third group, Rhipsalis, was by Stefan Savage. He began with the disclaimer that predicting way into the future is highly inaccurate. For instance, he noted that no one at HotOS-II would have guessed that the most significant problem of 1999 would be Y2K! His list of five problems were:
1.
Tools for building OSs

2.
Adaptive systems

3.
Science

4.
Device drivers

5.
Big groups of interconnected dumb devices

Savage observed that people doing OS design are not good at checking, and that this is a harder problem in OS design than in application design. This is why tools emerge as a leading requirement for OS research. Next, he explained that the need for adaptive systems arises because systems need to be robust across time, even in the face of changing specifications. He emphasized that the robustness had to be not just with respect to performance but also error handling and configuration. On the third problem, Savage characterized the OS community as being rather poor scientists. He advocated the goal of practicing the scientific method by testing hypotheses, rather than building systems and showing that they worked. This would require OS researchers to be better educated about statistics. Turning to the fourth problem, Savage noted that device drivers are not glamorous and no one publishes papers about them, but they are really important. Writing device drivers is difficult, but we don't fully understands why this is the case. Finally, he justified the last problem by observing the we are entering an era where the Internet and active ``stuff'' enters the home, and where each home has its own network. Many of the devices on such a network will not be smart enough to run real OSs, but there will be many problems in configuring them.


next up previous
Next: Rebutia Up: Sessions 7 & 8: Previous: Saguaro
Peter Druschel
1999-07-28