[c-nsp] Are Nexus and per-interface or FEX MTU settings possible?

Reuben Farrelly reuben-cisco-nsp at reub.net
Thu Sep 20 02:20:40 EDT 2012


I'd like to further clarify this - as I think the subtleties here 
between layer 2 and layer 3 MTU may be giving a misleading picture.

The Layer 2 MTU (AKA frame size) is set globally, in the same way as on 
a Catalyst floor switch such as a 3560/3750 is done.

This is performed by something like this in the config:

-------------

policy-map type network-qos enable-jumbo-frames
   class type network-qos class-default
     mtu 9216

system qos
   service-policy type network-qos enable-jumbo-frames

--------------

However the Layer 3 MTU is customisable, per VLAN interface:

---------------

SWT01(config)# int vlan 22
SWT01(config-if)# mtu ?
   <64-9216>  MTU size in bytes

SWT01(config-if)#

interface Vlan22
   no shutdown
   mtu 9216

-----------------------------

In the majority of cases you will probably want to have a high layer 2 
MTU globally, and then perhaps be more careful with your layer 3 MTU 
settings at your layer 3 boundaries (the default is almost always 1500 
so unless you specifically modify it, that's what it'll be, regardless 
of your layer 2 MTU).  If you aren't doing any Layer 3 switching on the 
device then the layer 3 MTU setting obviously won't be necessary on there.

The config above is from a 5548UP running 5.2(1)N1 (that has the L3 
daughter card).

I'm not aware of any negative side effects on this platform from setting 
the layer 2 MTU to a high value globally.  Things most certainly break 
badly when the L2 MTU is set too small, but not usually when it's too big.

Reuben


On 20/09/2012 4:01 PM, Lars Christensen wrote:
> Hi Joshua,
>
> You can't set a per interface MTU on the Nexus 5500-series switches.
> However, you can do it on per Class of Service.
>
> So in order to do increased MTU for your iSCSI traffic, you should
> create a policy-map identifying your iSCSI traffic, place the traffic
> in a separate queue with increased MTU.
>
> This is the way to go forward.
>
>
> Lars Christensen CCIE #20292


More information about the cisco-nsp mailing list