[c-nsp] 3750 running jumbo frames ?

[Peter Rathlev]

On Thu, 2009-06-11 at 09:44 -0400, Jeff Fitzwater wrote:
> We have the need to run two 3750 switches with jumbo frames (9000),  
> for a high performance data transfer application.  Both switches will  
> be manages by connections to a NON-JUMBO frame environment.   (That  
> is, if this will work)
> 
> If I enable jumbo frames (which is a global change) and leave the  
> management interface MTU at 1500 so the switch will use 1500 as packet  
> size for all management, is there any NEGATIVE ISSUES I should be  
> aware because of them being connected to the non-jumbo environment?

This will not present problems. As David mentions only L2 switched
frames can be jumbo. Management-traffic wouldn't exceed the routing MTU,
which is 1500 bytes by default. Changing the "system jumbo mtu" doesn't
change the L3 MTU.

Any TCP based L3 connection would use the lowest of the two endpoint
MSSs anyway, so hosts connecting from 1500 byte MTU segments would
always end up using 1500 byte MTU connections. Even if you could adjust
routing MTU to 9000 bytes you probably wouldn't face any problems.

IMHO there would never be any negative effects from enabling 9000 bytes
MTU, unless of course you explicitely WANT to limit the MTU.

Regards,
Peter