I had always thought (incorrectly I guess) that that disparity was due to the load balancing being based off of a "relative load" and not strictly the displayed queries/hour (call it "absolute load" if you want). By relative load I mean the (resource use)/(resource limit) while absolute load is just the plain resource use. Differently powered machines will have different resource limits (CPU, memory, bandwidth factors) leading to different strains on different machines even if they are experiencing the same QPH. Does it make sense to use relative load rather then absolute load? Is that sort of what is happening with the bandwidth/thread limits throttling the load numbers on some machines on the Freenet? Or is it *all* due to the fscking transients as Matthew mentions?
/Mike > > From: > Ian Clarke <ian at locut.us> > Date: > Tue, 11 Mar 2003 14:26:18 -0800 > > >Looking at http://127.0.0.1:8888/servlet/nodeinfo/networking/loadstats I >am surprised at the range of node loads being reported, ranging from 124 >queries/hour to a whopping 26412 queries/hour! Overall, there is a >standard deviation of 6519 (don't ask how i calculated that, but it >didn't help my RSI), and an average of 4367 (which, incedentally, is >different to the global mean as calculated by my node of 3219..?) > >This huge standard deviation would suggest that the load-balancing >really isn't working too well - any ideas? > >Ian. > > _______________________________________________ devl mailing list devl at freenetproject.org http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl
