On November 07, 2003 10:14 pm, Andrew Rodland wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> The way I'm reading the code (admittedly, I'm not entirely familiar with
> it), when we have a search eventually succeed after receiving one or more
> QRs, only the time it took to succeed on the last node is fed to
> globalEstimator. So, globalEstimator isn't the time it will take us to
> eventually get a key, it's the time it will take us to sucessfully get a
> key. But the QR-compensation factor in estimate() is pSearchFailed *
> (tSearchFailed + requestFailTime), and requestFailTime is a guess for our
> current key from globalEstimator. 

requestFailTime is really the total time we think a request with this key will
take to succeed.  It includes all QR etc.
 
> The upshot of which is, it looks to me like we are currently
> underestimating
 the potential risk of routing to a node that's likely to
> QR, because we underestimate the long-term cost of QR, because we don't
> consider *all* of the time that it will take to re-run the request in the
> cost of QRing. 
> All this, of course, assumes that I'm understanding the code in
> NGRouting.java that reports a value to globalEstimator properly -- so
> someone should probably check me on that before we get into anything else.

Ed
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to