SunCEC2007 Day 1: 4th breakout: Future The Future of Reliability by Gary Combs and Richard Elling
This was my last breakout yesterday. But it was really interesting. The talk answered many questions about future direction, what Sun will do in the future to make systems more reliable. Interesting facts from the talk:
- Integrated circuits have the same reliability regardless of the size of the structure. Thus reducing chip count by higher integration increases the reliability.
- RAS is a trade-off business. Serviceability may reduce reliability. Example: When you need extra chips to do hot-swap, the chip count increases, thus the reliability decreases.
- There will be really cool stuff to protect servers against memory failures in Supernova-Class systems (systems with the Rock processor). Really, really cool stuff ...
- Chip count of SMP-Systems vastly decreased since UPA, thus increased reliability.
- Solaris 10 Fault Management Architecture has a nice real world effect. Much less cases in certain areas, because of automatic reactions by the Solaris 10 operating system (like memory page retirement
- Sun designs Servers in a different way, where other systems have up to 6 or 8 currents at their power supplies, Sun power supplies generate only two currents. The power supply gets much simpler, thus more reliable.
The talk of the both was really informative. Was a good closure of the conference day.