LinuxSA Mailing list archives

  From: Andrew Pullin <andrew@hotspurbgc.com.au>
  To  : linuxsa <linuxsa@linuxsa.org.au>
  Date: Mon, 14 Jul 2003 22:04:58 +1000

IPv6 and frame sizes (was: Anyone going to meeting with laptop with GbE?)

Hi All,
    All this talk about bigger MTUs and jumbo frames has got me to thinking
about IPv6 and how it is supposed to work sending packets etc. I have
recently finished a course on TCP/IP, and we only touched on this topic (1/2
of one lecture).

    As I understand, IPv6 will try to work out the smallest chunk to
fragment packets into prior to sending, and then fragment the data into
packets of this size and send it. The idea is that the processing time used
to do this calculation, and the time spent collecting this data from the
path (apparently 5% of the total packet transmission time), is significantly
less than each router defragmenting and refragmenting the data enroute for
each hop. This sounded a bit Irish to me, as I would have thought that once
the packets left the sending network in an IPv4 network, they would start
out at 1500 byte packets (being the MTU for normal Ethernet) and gradually
collapse into larger packets as they were processed by bigger faster routers
before being shrunk back to 1500 byte packets for delivery on the
destination network. This sounded faster to me because as they grew into
larger packets it would seem that more data with less overhead would be sent
in a more efficient manner than in a lot of much smaller chunks. Sort of
like the theory that a racing car fitted with a 5 TB disk pack has a faster
data transfer rate than a hard disk drive. Also a little like the discussion
here that bigger MTUs on faster networks are more efficient.

    I am also a little confused in that if at present the standard for LANs
is ethernet (with an MTU of 1500 bytes), then under IPv6 wouldn't that be
the smallest MTU along the route and so the packets would be this big?
Wouldn't that mean that if the average data block to be transferred was 10
times this size that under IPv6 there would be 10 times as many packets that
would have to be routed? On routing, if the path is worked out prior to
transmission, what happens if the path changes enroute for different packets
to that the calculation was done on? If the new path had a route with an MTU
too small, would the data transfer fail automatically? Under IPv4 the data
would be split up into smaller packets and sent on the new path. Finally,
under IPv6 the hop count is restricted to prevent endless loops (amongst
other things), and this is good, but it is restricted to a relatively low
number, like 16 or 20 (how quickly I forget). Surely much data would need
more hops than that to get to some places. I know that to get from this
machine I am writing this email on will take at least 3 hops to get to the
gateway then the router and then to the next connection, and having just
done a traceroute to netcraft in Adelaide, it is 13 hops. If IPv6 limits
hops to 20, then I would assume that would make it very difficult to get
from Brisbane to Europe.

    Anyhow people, there seem to be some knowlegeable people out there in
the previous discussion and I would appreciate it if some of them could
throw some light onto this topic for me. I don't really have, and I am not
likely to get much experience with IPv6 for a few years yet, but it would be
nice to know how it works. TIA.
    Cheers!
        Andrew.

-- 
LinuxSA WWW: http://www.linuxsa.org.au/ IRC: #linuxsa on irc.freenode.net
To unsubscribe from the LinuxSA list:
  mail linuxsa-request@linuxsa.org.au with "unsubscribe" as the subject