THROUGHPUT BETWEEN 2 PCs

[netops044]

I have two laptops [Windows 7] each connected on end of 20Mbps pipe.  The MTU of the NICs is 1500bytes. Maximum pingable data size end-to-end  is only 1472 bytes with DF-bit set. When I initiate a normal windows  based file transfer end-to-end the throughput is 20Mbps in a single  session which is really good.

However when I took some  wireshark logs at the server end I noticed something interesting.

     The  maximum segment size [MSS] observed for each PDU is 2920 bytes  [2*1460]. In fact when I remove the bandwidth capping on the last mile  the      single session speed increases to very good speeds of 70Mbps.

Observation  - 1) DF bit is set in the packets    2) The PC-to-PC maximum pingable  datagram size is 1472 with DF bit set    3) The message "TCP segment of a  reassembled PDU" is being observed for each & every segment     4)   Normal windows file sharing is used     5) Application is  "microsoft-ds" port number 445

My question is:  How can the MSS segment  size be 2*1460 bytes? I feel it should be 1460bytes only. Do share your  observations on the same. Attached is a snapshot of the a single packet  capture, single the whole file is too big

[Giuseppe Larosa]

Hello Netops44,

from the fact that MTU is only 1472 bytes I guess a form of tunneling may point-to-point GRE tunneling is in use of MTU would be 1500 bytes.

the WAN device may override the DF fragment in order to be able to fragment the GRE packet that carries the user packet inside, on the other end of the link the original packet is re-assembled including the DF bit set.

As a result of the scenario described above TCP MSS can be as high as the observed value as TCP is not aware of fragmentation that happens on the external GRE packet.

the DF bit can be cleared on the external GRE packet by using policy based routing with action

see

http://www.cisco.com/en/US/partner/tech/tk827/tk369/technologies_tech_note09186a0080093f1f.shtml

Hope to help

Giuseppe