By user12625760 on Oct 03, 2006
I'll leave the write ups of the main sessions to those who are competing for the best CEC blog. I could run the risk of being accused of “brown nosing” if I wrote what I thought.
I managed to get to three excellent brake out sessions the first of which was about SGRT. Specifically how to use SGRT when dealing with performance escalations. Clive King and Venkat Ramani specifically explained how to answer the questions:
“Where on the Object” and “Where in the life cycle” for performance problem during situation appraisal. The simplicity and power of the tools and concepts use were stunning.
For the “Where on the Object”, Clive demonstrated the difficulty of applying this question to a software system where you have nothing physical to point to or touch. He simply had a diagram of how IO is done in Solaris showing all the levels from the user application down to the disk drive. The original diagram is in the Solaris Internals book. Using this you get the possible places for the answer to the where on the object question. By then explaining how different tools will give you different timing points with in that system you can narrow down where the issue is on that diagram. So iostat(1M) can be used to see if the problem is below that target driver, typically sd(7D), and tnf probes to see if the problem is below the system call interface. Immediately you can narrow down the “Where on the Object” question to “above the system call interface” or “between the system call interface and the target driver” or “below the system call interface”. Using dtrace(1) on Solaris 10 you can dive in deeper.
All in all an enlightening talk.