[e2e] TCP Fragmentation

Cannara cannara at attglobal.net
Fri Jan 31 19:04:35 PST 2003


And, application/MAC configuration can easily affect this -- a local DOE lab
has servers thinking they're configured for FDDI, so sending 4kB PDUs down to
IP, which then does fragments.  Other protocols (AFS, NFS...) often simply
send Enet-sized IP PDUs as fragments.  Of course, fragmentation to smaller
MTUs than Enet should certainly not be around anymore -- ha, ha.

On the other side, setting the DF bit can be done with imaginary
FDDI/Token-Ring installs on Enet -- most Enet MACs can handle 4k frames (as
can bridges now), so efficiency goes up for transfers.

Alex

Janardhan Iyengar wrote:
> 
> John,
> 
> According to "Beyond Folklore: Observations on Fragmented Traffic" by
> Shannon, Moore and Claffy (from ToN, December 2002):
> 
> "While majority of the fragmented trafficis UDP [..], ICMP, IPSEC, TCP and
> tunneled traffic are commonly fragmented as well."
> 
> Although that doesn't answer your question per se, these results show that
> there are TCP stacks which allow fragmenting.
> 
> regards,
> jana
> 
> On Mon, 27 Jan 2003, John Border wrote:
> 
> >
> >     Is anyone aware of a TCP stack in use in the "real world" which does not
> > set the "don't fragment" bit in the IP header?
> >
> >
> > John
> >




More information about the end2end-interest mailing list