GRE Tunnel effect on overall bandwidth

[jgorman1977]

We have a 6Mb MPLS connection from a branch office to our core office. I had to create GRE tunnels as the default route was not being passed through the cloud.

My question is, how will the GRE tunnel effect the overall throughput?

Here is the serial interface output:

Serial1/0 is up, line protocol is up

Hardware is DSXPNM Serial

Internet address is 172.40.0.10/30

MTU 4470 bytes, BW 6015 Kbit, DLY 200 usec,

reliability 255/255, txload 14/255, rxload 29/255

Encapsulation HDLC, crc 16, loopback not set

Here is the Tunnel interface output:

router-nj01#sh interfaces tunnel 2

Tunnel2 is up, line protocol is up

Hardware is Tunnel

Description: Tunnel to Indianapolis

Internet address is 192.168.150.6/30

MTU 1514 bytes, BW 9 Kbit, DLY 500000 usec,

reliability 255/255, txload 26/255, rxload 1/255

Encapsulation TUNNEL, loopback not set

Thank you in advance

[Edison Ortiz]

My question is, how will the GRE tunnel effect the overall throughput?

GRE will increase the size of the packet so you must be aware of fragmentation issues on the transit path.

Here is an example of how GRE can affect your WAN connection:

http://www.cisco.com/en/US/tech/tk827/tk369/technologies_tech_note09186a0080093f1f.shtml

HTH,

__

Edison.

[jgorman1977]

Edison,

Looking at the MTU of 1514 on my Tunnel interface, would bumping it up to 1524 be a solution?

Thanks

[Edison Ortiz]

You would assume that every single device in the path would support that packet size? I don't think that's possible.

Best to conform the packet size standard of 1500 max (including the GRE header). The link I provided has many ways of working around this issue.

__

Edison.

[Mohamed Sobair]

Hi,

Adding to Edison's:

GRE adds additional 4byte header to the 20 Byte IP header. Once GRE is configured, and to prevent performance degredation caused by Fragmentation, always check the MTUs.

The link provided by Edison must be very useful.

HTH

Mohamed

[foxbatreco]

Hi,

Interesting discussion..just to check out some of my queries.Does GRE has or supports any mechanism for symmetrical path traffic.Eg: Site A sends out loadbalanced traffic by ospf through two upstream provider routers & has 2 GRE tunnel to the other end.Will the return traffic back to site A follow the same path as it went out(either link1 or link2)or once it reaches within PE mpls cloud...it would just end up using more of one of the links?Thanks.

[Joseph W. Doherty]

If you're running OSPF across 2 GRE tunnels as you describe, then OSPF should route across them as it would with any other p-2-p links. If they have different cots, OSPF should take the less expensive path. If they have equal costs, OSPF should use both. Return traffic should be an independent forwarding decision. I.e., traffic might, or might not, return on the same GRE tunnel.

[foxbatreco]

Ye..both have equal costs.return path is the one we are looking at..anyway to influence this so that session flow can be along the same path as it went outbound.

Thanks!

[Joseph W. Doherty]

"anyway to influence this so that session flow can be along the same path as it went outbound."

If you mean some kind of dynamic "stickyness", i.e. inbound flow uses same path as outbound did, don't believe so.

Statically, you could either stop using equal cost with OSPF, and/or use something like PBR.