> > what about basic pcache not on "first response time" (latency) but on
"transfer speed" (ticker started when key is requested and stopped when the
last byte of
> > the requested key is tranferred, resulting into a bytes/sec value, just
like on the splitfile information pane) ?
> > in fact, everything what's important is the time needed to have this key
on our side, not the latency or other things, so it would make sense to
actually respect
> > that value ;)
> >
> >
>
> The tricky part is that there are 32KB files mixed in with 1 MB files
> along the keyspace.  Mixing results from different file sizes screw up
> the RTE vectors because 1 MB files take significantly longer to download
> than 32 KB files.  One solution might be making the RTE vectors 3
> dimensional with the file size as the third dimension.

True, but I'm not sure this hurts so much if you divide it out to be
bytes/sec. The overhead costs would end up counting less for bigger files
than smaller ones, but unless there's some real wackiness in the
distribution of sizes of requested keys, I don't think that it would make a
significant difference. But it is something to consider.

You could, of course, just subtract out the setup time, but then you're
treating the smaller keys unfairly, instead of the bigger ones.

Could you illuminate me (us) on the mathematics of a proposed 3d impl?

--hobbs

_______________________________________________
devl mailing list
[EMAIL PROTECTED]
http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl

Reply via email to