[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: GSE IDs [Re: IETF multihoming powder: just add IPv6 and stir]

On dinsdag, mei 13, 2003, at 13:40 Europe/Amsterdam, marcelo bagnulo wrote:

But I think we should revisit this when there is an actual proposal
that adds extra headers to packets because only then we'll be able to
see how this helps us and how it hurts us.



Well, using MIPv6 could be one of such proposals (there are some issues
to solve) but still can be useful to quantify the amount of overhead
involved.




Mobile IP in IPv6 uses a 24 byte header to carry the original source 
address, right?



If we assume ethernet with a 1500 byte MTU and include the ethernet 
encapsulation (MAC addresses, ethertype and frame check sequence) but 
not the lower layer stuff (this is a good compromise between looking 
only at IP, which is slightly worse or also including the extra framing 
bits/gap, which is another 20 bytes per packet) and assume full sized 
packets we arrive at:



              IPv4   IPv6   MIPv6

Data          1460   1440   1416

Overhead        58     78    112

Efficiency1   96.2%  94.9%  93.3%

Efficiency2   94.4%  92.5%  90.3%



In the efficiency1 figure we drop 2.9% when moving from IPv4 to MIPv6 
but this figure doesn't include the TCP ACKs that nearly double in 
size. For a full duplex link this adds up to another 3% in overhead 
while this is only 1.8% for IPv4 so the overall drop in efficiency is 
4.1%.



Now if you're going to do high speed data transfers you'll obviously 
want to enable the RFC 1323 high performance extensions which means 
another 12 bytes per packet are used for timestamp landing you at 89.1% 
efficiency (93.2% in IPv4). And then there is IPsec...



In the real world all of this is a lot worse because the average packet 
size is significantly smaller than the maximum possible packet size.



Now all of this may not seem like a huge deal but what if the bank 
decides they want to take 4% of the money in your account, or look at 
the trouble people go through to lose 4% of their weight. For some 
people this may be a reason to stick with IPv4. On the other hand, if 
we can keep it at 4% it might be ok. But if the application protocols 
(which usually have HUGE overhead) and all the lower layers add 4% too 
then it's no fun anymore.