> > There have been a number of really awsome tools that > sadly haven't been > marketed or taught to customers very well; they would > include, > kstat > microstate accounting
I hear you on the microstate accounting. It amazes me how many people install top on Solaris >= 8, when prstat is substantially better... It's also quite nice that the data is accessible via procfs data structures. > The "Solaris Performance and Tools" book does demo a > lot of these, and > mentions kstat frequently (including sample > Sun::Solaris::Kstat scripts in > the appendix). Hopefully it will promote some of > these lesser known tools. As far as kstat goes, one thing I've been wishing for a long time is for Sun to release the Sun::Solaris::Kstat module in source form to CPAN (or through a Sun website). It's a pretty significant obstacle to realistic use that the Sun::Solaris::Kstat module may, in my experience, not be binary compatible with other perl installations, and that Sun's documentation specifically discourages addition of external site_perl modules to the Sun-provided /usr/perl5 perl installation. This makes it difficult to use kstat data through the perl bindings, which are significantly preferable to the native-code API for system administrator/system engineer folks doing lighter-weight tools work. This thread in general is kind of funny to me, since in my view, Linux's internal accounting facilities are actually incredibly weak compared to Solaris. For example, Linux statistics gathered through various procfs pseudofiles like /proc/stat, /proc/PID/stat, etc. don't bother to include an analogue to a kstat member's snaptime. Without that, it's pretty difficult to get accurate time deltas. If anything, I generally find myself wishing Linux provided the kind of data and structured interfaces available on Solaris. -- This messages posted from opensolaris.org
