[mdlug] OT: the great IPv6 debate

[Dan Pritts]

On Wed, Apr 21, 2010 at 06:28:41AM -0400, Adam Tauno Williams wrote:
> Not to mention: better performance.  Simple headers, no fragmentation,
> no per-hop checksum, much larger MTUs, 
 
In practice, MTUs are a layer-2 issue, not a layer-3 issue.   

yes, ipv6 specifies a larger minimum MTU (1280 vs 500-something),

And yes, ipv4 is limited to 64k MTU while ipv6 doesn't really have
a limit (right?).

but in practice nearly everyone uses standard ethernet's 1500, or
1492 if they're stuck behind PPPOE.  In the data center, or in HPC
transfers over WANs, 9000 bytes is common.

Bigger MTUs are great, don't get me wrong, but ipv4 can handle an MTU
7x as big as the current standard "jumbo frame" of 9000 bytes.

> and ***much*** smaller routing
> tables provide an enormous load reduction for routers.  [The IPv4

those small routing tables are why multihoming is broken.  

Don't get me wrong, it's possible to design around the lack of
multihoming at higher layers, but it's not simple and that work has
not yet been done.   And there's something to be said to moving the
complexity out of the network and into the edge; it's more scalable
that way, for sure.

danno
--
dan pritts
danno at umich.edu
734-929-9770