hi Andreas,
It's actually of interest to me, how much endianness affects the ipv6 processing in either a router or end device. But I couldn't find any benchmarks that would compare the same CPU in BE and LE mode under the same Linux kernel version. The only references that I found told it's faster in BE. Also by looking at the kernel sources, I clearly see where exactly it's faster.
I'll probably have to do the benchmarking myself, as soon as I find a board that is easy to switch between LE and BE modes.
Hardware support of checksum calculation is another topic, quite interesting, but way out of software geek's control :) Although it's possible to use FPGA for such operations... which leads to another interesting project :)
________________________________ From: Andreas Fink afink@list.fink.org To: Stanislav Sinyagin ssinyagin@yahoo.com Cc: swinog@swinog.ch Sent: Tuesday, March 3, 2009 6:15:32 AM Subject: Re: [swinog] IPV6 Go (lazy providers)
Byteswapping of addresses and netmasks takes like a nanosecond on the systems which require swapping. So dont waste your time on that. CRC checking is way more CPU intensive on TCP but that's done nowadays in hardware on the ethernet card on modern systems and its the same for IPv4 and IPv6.