"Too large frames" - what does this mean for performance?

[oddarneffk]

Hi!

We have a problem with a switch connected to an IBM Chassis. The links to the chassis seem to flap from time to time and I find this in the logs:

Transmit GigabitEthernet1/0/15           Receive
641365586 Too large frames

Why are there so many too large frames? We have not done any changes to any mtu size so it should all be at 1500. Could it be the server is sending too large frames and the switch drops them, or is it the switch that drops frames bound for the ibm chassis that are too large?

[Bert Gevers · Accepted solution]

Hi,

the counter 'Too large frames' counts the total number of frames transmitted whose wire lenght ( including FCS) is greater than 1518 bytes.

On the receive side, you will have a similar counter being: valid frames, too large

Basically, these counters may increment during normal operation on a trunk link (due to the addition of the

802.1Q header or ISL headers) .  These counters do not indicate that these packets are being dropped.

Please note that this behavior may differ upon the software version you run.
Eg. Catalysts 3750/3560s running IOS version 12.2(44)SE2 or later will increment these counters.  The same devices running older IOS versions will not increment these counters.

Best regards,

Bert

[oddarneffk]

Aha!

I thought that these counters showed packets that were dropped.

What happens when a switch receives a packet that is > 1500 bytes? Does it drop it, or does it split it into smaller <= 1500 packets and

then forward them?

[Edit: Typo: Changed "splice" to "split"]

[Bert Gevers · Accepted solution]

For valid frames (eg. traffic on trunks), the switch will attempt to forward them.

The switch will not perform fragmentation on these packets.

Bert

[oddarneffk]

..ok, so in other words it drops frames that are bigger than the MTU, UNLESS it's a trunk port where it does it's best to forward such packets.

How can I tell that packets have been dropped due to their size > MTU ?

[Bert Gevers · Accepted solution]

Hi again,

The dropped packets will be seen as giants on the interface counters for the 3560/3750.

Eg.

- Frames over 1518 bytes increment the 'Valid frames, too large' counter and are forwarded.
- Frames over the valid system/routing MTU increment also the 'Valid oversize frames' and 
'giants' counter in 'sh interface' and are dropped.

You can use 'giant' counter in 'sh interface' to troubleshoot any MTU mismatch issues.
I hope this answers your question.

Bert

[oddarneffk]

Yes it did. Thanks a lot

[gnijs]

Hello Bert,

I would like to go a little further in this explanation, and especially the behaviour on trunk ports.

Suppose i am sending a max. sized packet ie 1500 bytes data, 1518 on the wire.

This packet hits a L2 Portchannel and a 802.1q header needs to be added.

Is this packet now counted as an "Valid frames, too large" or "Too large frames".

From what i am seeing on a C3120G running 12.2(50)SE3, this counter is incremented (rapidly), although i receive no complaints.

So i guess there are no drops. I also see no giants errors.

Is this normal behaviour ?

If i would increase the system/routingMTU, would these counters stop to increase ?

regards,

Geert

[Bert Gevers]

Hi Geert,

if the sending side is the 3750/2560, it will report them as 'too large frames'.  The receiving side should report 'valid frames, too large'.

This is normal behavior.  There is no impact due to this counters (eg. the frames are not dropped) and changing the MTU will not make a difference on the device as this will only affect the 'valid oversize frame' counter.

HTH,

Bert

[ganderson]

Thanks Bert! This makes perfect sense.Now that I understand why I should worry about it I won't. 

Have a great day.