NANOG 08 feb 98 1400 kc presentatin




NLANR cache hierarchy statistics,
visualization, and roi


8 feb 98


nlanr cache pi: Duane Wessels
wessels@nlanr.net
data analysis: Alex Rousskov
rousskov@nlanr.net









aggregate statistics


daily traffic per NLANR caches (bytes and documents)












daily traffic NLANR supercache

11 NLANR caches, off and on: BO, CH, DC, IT, LJ, PA, PB, SD, SJ, SV, UC.












cumulative traffic ( 1 webit = O(10TB) )














savings (doc, byte hit ratios) over course of project

supercache daily savings


absolute daily savings

cumulative savings


cumulative supercache savings (almost 1 webit)

note superlinear growth of saving rate.










hierarchy statistics


(more recently collected) hierarchy analysis

direct fetch, remote miss

local hits, remote hits

Note SJ cache has no remote hits and misses.



per-type statistics


absolute traffic (BO, (PB's stats similar)

doc types of 5% or more of intensity or volume

images largest category
but which documents most cache-valuable?

relative contribution per traffic category

per-type savings (images well ahead)

cumulative per-type savings

executables in 2nd place,
but low hit ratio reduces their caching efficacy


visualization

return on investment


9 feb 98, alex/kc, info@caida.org.