Sean Liu wrote: > Well, if we lost the clue of freemem 30 years ago, where the heck did the > vmstat freemem Solaris 8 - 10 w/o ZFS come from??? Out of nowhere? > > And subtract zfs:arcstats:c_min from what? > > If I am asking the wrong question, please give me the right question. >
I try to expose some of the calculations in arc_summary (http://www.cuddletech.com/arc_summary). Using that tool you can observe how much bang for your buck your getting inside of ARC. Feel free to rip the calculation out of it. I think the suggestion was this: zfs:0:arcstats:size represents the current size of the ARC zfs:0:arcstats:c_min represents the minimum size of the ARC, what it can shrink to. Therefore, size - c_min == the amount of used but potentially free-able memory. I'll note that in my monitoring I pull unix:0:system_pages:pp_kernel, which includes the size of ARC, and then subtract out zfs:0:arcstats:size so that I can see Kernel and ARC as seperate values. I would agree with your fundamental point that ARC to some degree should be considered free in the classical sense to be constant with what administrators expect from our UFS heritage. benr. _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
