08-30-2011 01:27 AM
For a pilot phase I've set up a new Juniper infrastructure including two mixed VC (2x4500, 2x4200) as core switches and two access switches (EX2200, Cisco 3560). We fixed the MTU to 1500 on all Uplinks and AEs because the customer wants to do a 1:1 replacement.
When we connected the mentioned setup to the existing Cisco core switch (6500 trunk port) everything was working fine with small sized frames. When I was connected on the Juniper access switch, I was able to ping something routed on the Cisco core with the maximum frame size of 1450 bytes.
On the existing infrastructure pings are working upto 65535 bytes. The are fragmented of course. But why does this not work through the Juniper infrastructure? I guess the EX devices are able to fragment the traffic automatically. Is there a mismatch between the Cisco MTU 1500 and the Juniper MTU 1500?
Thanks in advance for your answers,
08-30-2011 04:33 AM
In IOS Cisco accounts for L2 overhead automatically when you code MTU on an interface..
On a L2 trunk on IOS when you set MTU 1500, it really means 1518 (Dot1Q and Ethernet overhead)
On a L2 access on IOS when you set MTU 1500, it really means 1514 (Ethernet overhead)
On a L3 interface on IOS when you set MTU 1500, it really means 1514 (Ethernet overhead)
To get JUNOS to be equivalient to IOS you have to take in account for the medium overhead, in your case being ethernet you need to add the bytes above as mentioned.
09-01-2011 03:06 AM
just a quick correction:
the maximum frame size for standard ethernet is 1518 byte, in case you are doing vlan-tagging on a trunk port (802.1q) the max. frame size is increased to 1522 byte.