On Thu, Apr 08, 2010 at 10:36:15AM -0700, John Plocher wrote: > I presume it is a good thing to have so many stats, and to be able to > inundate the admin with tons of info, but what is missing for me is > the conceptual structure to interpret and utilize all this info and to > actually debug problems. In ARC-speak, what (where?) is the > user/usage model that pulls all these stats together into a coherent > whole? Or, do I need to be an expert CIFS protocol/implementation > expert to make sense of any of this?
You might need to write DTrace scripts to debug performance problems on your own. And if you're using unstable probes then you must have knowledge of the relevant implementation and/or protocol details. IMO the stats themselves, both kstats and kernel symbols/variables that'd be meaningful to refer to from DTrace scripts should be listed as Volatile, as that will allow for programmatic access with the necessary "could break between releases" caveat. But that could come later. Nico -- _______________________________________________ opensolaris-arc mailing list [email protected]
