SUP720 and GRE

Konstantin Dunaev · ‎01-09-2008

Hi,

we have some strange situation.

According to the documentation the GRE traffic on SUP720 should be processed in CEF mode. But If I look onto our graps I can clearly see that there is a dependency between amount of GRE traffic and management CPU load: GRE traffic is about 400Mbit and COPU load is about 20%.

I tried to "sniff" the traffic which is procesed by CPU (some strange "remote" command) but I don't see any reason why that traffic goes throug the CPU - it's not fragmented, the size of packets are less then tunnel MTU size.

Which traffic on SUP720 are not processed in CEF mode and sent to CPU?

Edison Ortiz · ‎01-09-2008

Are you using the same source IP address on different GRE tunnels?

__

Edison.

Konstantin Dunaev · ‎01-10-2008

hi Edison,

that was my first thought, but I checked the configurtion and no, only one GRE tunnel is active and it use the loopback interface as source:

interface Loopback0

ip address y.y.y.y

!

interface Tunnel11

ip address x.x.x.x

ip mtu 1450

keepalive 10 3

tunnel source Loopback0

tunnel destination z.z.z.z

tunnel path-mtu-discovery

Edison Ortiz · ‎01-10-2008

You mentioned you aren't fragmenting the packet but I see the mtu set to 1450. You will definitely encounter packets greater than that as the default on ethernet is 1500, unless you have an intermediate device fragmenting the packet for your before it gets to this device, packet fragmentation will occur.

This may be the reason for the CPU hit, fragmented packets go in Software, not Hardware.

Can you also post the IOS version?

__

Edison.

Konstantin Dunaev · ‎01-10-2008

Hi,

yes we're using IP MTU 1450 on GRE tunnel and PMTUD protocol works fine, I see that the size of packets originating from the source hosts, which generate the most load on GRE tunnel, is not more then 1414 bytes (if I sniff directly on those hosts) and a lot of "TCP window update" (66 bytes), but on the sniff from SUP720's CPU I have a lot of packets with size of 70 bytes (4 bytes for GRE header?) but not all, eg. directly on host I can see ~10 TCP updates packets, and on SUP720 only 2 or 1 from those 10.

IOS 12.2(18)SXF7, RELEASE SOFTWARE (fc1)

How can I reduse the number of generated "TCP update packets"?

Edison Ortiz · ‎01-10-2008

The 6500 is telling those hosts to reduce the packet size via ICMP. As a test, try to create a flow between two end-devices on the same segment. Take a trace and you will see 1500 sized packets.

Perhaps what you are seeing is the host reducing the packet size to fit on the GRE tunnel. I wonder if this process is causing a CPU load.

__

Edison.

Konstantin Dunaev · ‎01-10-2008

>Perhaps what you are seeing is the host

>reducing the packet size to fit on the GRE

>tunnel. I wonder if this process is causing

>a CPU load.

it seems that the "TCP update" are processed by CPU on SUP720. because I got a lot of them.