agenda item suggestion for Thursday: kernel debugging and performance tuning

[Vara Prasad]

Hi,

Tomorrow we have a mtg at IBM that we all have to leave at 10 AM,
hence it is going to be a little short meeting, sorry for this change.

bye,
Vara Prasad

Chen, Brad wrote:

>I was hoping we could establish consensus on a few notions 
>and priorities.
>
>- that we want to build a great tool for both
>  performance tuning and for kernel debugging
>- special needs for kernel debugging include
>  - must be able to instrument anywhere
>  - must be able to read or write any kernel data
>  - requires minimum restrictions on kernel routines
>    that can be invoked from a tap
>  - other needs?
>
>- special needs for performance tool users include:
>  - safe: systemtap is broken if it is possible to
>    crash or damage the system
>  - fast: many useful tools have overhead < 5%
>  - easy to use: can do many useful things without the
>    need to read kernel source. I can run scripts from
>    3rd parties without fear.
>  - need timer events, stack walk into user space, tap
>    sets for key OS abstractions: system calls, I/O,
>    synch, scheduler, network, ...
>  - other needs?
>
>Questions: Do we agree we want to try to meet the needs
>of both kinds of users? What kind of options or auxilliary
>components can we use to give these substantially different 
>perspectives for the two groups of users? When there is a 
>conflict, do we prioritize needs of one kind of user over 
>the other?
>
>Brad
>
>  
>