Oskar wondered:
> I don't remember how big UDP datagrams can be, but if they can fit a
> whole request routing message (not a data message), then a UDP based
> system would certainly be advantageous for Request performance.
According to:
http://lib.nevalink.ru/tcp_stevens/udp_user.htm#11_10
....you can theoretically include up to 65507 bytes of user data --
65535 bytes in the datagram, minus 20 bytes of IP header, minus 8 bytes
of UDP header. In practice, however, data that big is likely to get
truncated(!) or just plain dropped. The author suggests that 8192 bytes
is the biggest reliable size. I think that includes the 28 bytes of
header info, so that leaves 8164 bytes of user data per UDP packet.
If and when we ever get around to implementing UDP, we may want to have
the system try bigger packets and/or the "don't fragment" flag, to
improve performance between the links that can handle it. We may have
to implement a "datagram size throttle" anyway, if we find that even
the 8192-byte datagrams are getting dropped too much.
I'm not sure where UDP should fit on the road map. Certainly we should
not bother with it for a while. UDP has possiblities in the realms of
steganography and resistance to traffic-analysys, but it looks to be
mostly a performance thing. Some weekend warrior with a good UDP
background may be able to tack it on in one day, so who knows when it
will happen.
--Will
(never speaking for his employers)
willdye at willdye.com
P.S. FWIW, the raw RFC can be found at:
http://www.cis.ohio-state.edu/htbin/rfc/rfc768.html
_______________________________________________
Freenet-dev mailing list
Freenet-dev at lists.sourceforge.net
http://lists.sourceforge.net/mailman/listinfo/freenet-dev
