-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Matthew Toseland wrote:
> No, at present we back off if a node rejects a request or times out. We
> reject requests if our ping time is too high, to prevent timeouts.

OK, so instead of using the average ping time across all peers, should
we subtract our own measurement of the per-peer ping time from the
peer's measurement of the per-peer ping time, *then* average across all
peers, and if the average is large and positive, reject requests?

By subtracting our own measurement from the peer's measurement, the
network round-trip time cancels out, so instead of measuring the network
delay (a poor indicator of load) we're measuring the processing delay.
If our own processing delay is higher than that of most of our peers, we
conclude that we're overloaded.

Cheers,
Michael
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)

iD8DBQFE6zUYyua14OQlJ3sRAiAcAKDxOh3cUZeH1P/Ep14aDDEzkEuhZwCfb09f
HxUZa+a+Ds6Vd6iJL0h2XY8=
=uEGP
-----END PGP SIGNATURE-----

Reply via email to