How are the Routers GRE capability

[suguoqiang]

Dear Experts，

      I want to ask about each router's GRE performance, how much is the GRE throughput? Can you give me some offical document link? Thanks a lot

[Edison Ortiz]

The throughput is regulated by the transit links that are used by the GRE tunnel, not the tunnel on itself.

For instance, if you have a 3845 ISR router with a DS3 connected to a remote router with the same capabilities,

you can potentially send line rate speed throughput via a GRE tunnel.

Please note, I said 'potentially' as other configuration issues must be addressed at the client side and the router to

address any MTU issues the GRE tunnel brings into the picture.

Regards,

Edison

[Peter Paluch]

Edison,

I am glad you responded. Having read your answer, one question pops on my mind: Are the GRE-tunelled packets CEF switched, or are they punted and processed by the CPU? I am afraid that if all GRE-tunelled traffic (incoming or outgoing) must be processed by the CPU, it will be the main bottleneck.

Best regards,

Peter

[marikakis]

Hi Peter,

I am also glad you asked this question. What one means by 'throughput' depends. One can refer to some type of rate on the line in terms of packet overheads or use it to refer to packet switching performance. In any case, the packet switching performance will affect the rate observed on the line. The actual behavior is probably platform dependent. See for example the following links:

http://www.cisco.com/en/US/docs/ios/12_0s/feature/guide/gre.html#wp1031648

http://www.gossamer-threads.com/lists/cisco/nsp/106183

So, maybe the original question needs to be narrowed down to specific platforms. Unless Edison has some general document that can share with us.

Kind Regards,

Maria

[Edison Ortiz]

Hello Peter,

As Maria indicated, the answer depends on the platform of use. However, since we are in the routing and switching section of NetPro, my answer was more inclined to the VXR/ISR routers and packets traversing a tunnel interface are CEF switched.

Rack1R1#sh ver | i IOS
Cisco IOS Software, 2800 Software (C2800NM-ADVENTERPRISEK9-M), Version 12.4(15)T9, RELEASE SOFTWARE (fc5)
!

!

Rack1R1#sh ip route 10.26.6.6 | i Tunnel
  Last update from 10.255.255.2 on Tunnel0, 2w2d ago
  * 10.255.255.2, from 10.255.255.2, 2w2d ago, via Tunnel0

!

!
Rack1R1#sh ip cef 10.26.6.6 det
10.26.6.6/32, version 32, epoch 0
0 packets, 0 bytes
  via 10.255.255.2, Tunnel0, 0 dependencies
    next hop 10.255.255.2, Tunnel0
    valid adjacency

If this flow was punted to the CPU, I would see valid punt adjancency

http://www.cisco.com/en/US/products/sw/iosswrel/ps1828/products_tech_note09186a00801e1e46.shtml

Regards

Edison

[Peter Paluch]

Maria, Edison,

Thanks to both of you for your kind insights and answers.

Best regards,

Peter

[suguoqiang]

Dear experts,

        Actually I want to ask what is the transmission rate via GRE tunnel of each hardware platform, such as ISR 3845 or VXR 7206 7301, the router is perform as tunnel source. Is there any document describe the maximum GRE throughput, the router generate the tunnel packet by itself.

[Edison Ortiz]

I believe I addressed your concern but you want a document to back it up?

https://www.cisco.com/en/US/docs/solutions/Enterprise/WAN_and_MAN/P2P_GRE_IPSec/4_p2pGRE_Phase2.html

If the document above is what you need, please mark this thread as resolved.

Regards

Edison

[suguoqiang]

Dear Edison,

          Thanks for your answer first, I've looked through the document, but what it say is about GRE over IPSec, but in our case we don't need the IPSec encryption, we only need tunnelled the packect with private network address, i think IPSec encryption may lead to much higher CPU usage.

[suguoqiang]

Dear experts,

       Please give me some help on the issue. Thank you so much.

[Jon Marshall]

suguoqiang wrote:
Dear experts,
       Please give me some help on the issue. Thank you so much.

Not entirely sure what more can be said. As Edison clearly pointed out the limiting factor will be the throughput of the router and whether GRE is CEF or non/CEF switched.

Attached is a pdf which shows router performance for process switched and fast switched/CEF packets. It doesn't specifically mention GRE but it does give the throughput of each router.

Jon

[marikakis]

Hi all,

In the last document posted by Edison there exists a table 4-2 for the p2p GRE only case:

https://www.cisco.com/en/US/docs/solutions/Enterprise/WAN_and_MAN/P2P_GRE_IPSec/4_p2pGRE_Phase2.html#wp91542

although it only shows results for some of the platforms. For the IPsec cases there is some hardware type assistance, so maybe you can't say this the worst expected performance in the GRE only case. However, maybe you can set some rough limits for the other platforms by seeing approximate results for higher end platforms in the GRE only case.

Jon, it seems acknowledged experts always have something more to say, since the document you posted is very interesting. I especially liked the text with the warnings before reading Table 1. Questions like 'how many DSL customers per BRAS? how many OSPF neighbors per router? how many areas attached per ABR? how many BGP sessions per router? etc" always boil down to the answer "it depends". The most authoritative expert to answer such questions is the router. You load the router with the features and the router spontaneously responds.

Kind Regards,

Maria