On Fri, Feb 13, 2009 at 04:53:18PM -0500, Richard Lowe wrote: > Jason King <ja...@ansipunx.net> writes: > > > On Fri, Feb 13, 2009 at 3:12 PM, Brendan Gregg - Sun Microsystems > > <bren...@sun.com> wrote: > >> On Fri, Feb 13, 2009 at 08:09:52PM +0000, Peter Tribble wrote: > >>> On Fri, Feb 13, 2009 at 12:26 AM, Brendan Gregg - Sun Microsystems > >>> <bren...@sun.com> wrote: > >>> > > >>> > No. Stop. Do not assume any data is better than no data. Wrong or > >>> > misleading > >>> > data is *worse* than no data. > >>> > >>> Well, that's not *entirely* true. Senior management isn't satisfied > >>> with no data, > >>> and prefer simplistic data that they're familiar with - even if we > >>> know that it's > >>> wrong or misleading. > >> > >> I think the problem here is one of thinking like an end-user, rather than > >> an engineer. In OpenSolaris, we can engineer whatever is needed - we don't > >> need to make-do with what engineers give us - we are the engineers. > > > > I would like the think that, all the (summarized) 'never use any > > kstats -- those are private' emails I'm getting off list, as well as > > past reactions I've seen seem to suggest otherwise (not that I'm > > really going to let it stop things -- I'd rather have a good tool, > > even if it has to be unbundled because *gasp* it might happen to use > > say 20 different kstats and Sun won't allow something that uses kstats > > that they didn't write to be putback in any consolidation). > > I think this has to be based on a misunderstanding on somebody's part, > though I don't know whose.
Yes, the message is not to never use kstats. We can use kstats, but we need to be careful and understand that they currently don't have stability semantics, and as such are subject to change at any point (which makes them dangerous when you want to export them via a documented SNMP MIB.) However my point in particular was to not assume that kstats (or vmstat numbers) are the Right Metrics to use simply because they are there, but to think about what metrics are actually needed, and *then* see which kstats satisfy them or which need to be added. Brendan -- Brendan Gregg, Sun Microsystems Fishworks. http://blogs.sun.com/brendan _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
