We have Mikrotik RB1100's doing around 400,000PPS happily with a fairly even mix of frame sizes.   They really are at their upper limit though....
We have RouterOS on Dell R210's doing ~7.2gbit throughput through VPLS tunnels, unsure of the PPS but CPU load averages around 40%

There is a new RB1100AHx2 that will apparently do 1million PPS.


The gear is cheap enough that you can buy a few to test your particular scenario on.

If it will help ease your mind I can give a demo of the Q-in-Q/VPLS capabilities of RouterOS .  One big benefit is that they have an excellent API and scripting so you  can easily script a new customer e.g. create parent vlan, create sub vlan, apply mangle policy to sub vlan, create queue instance for traffic policing.




Regards,





Andrew




On 22/12/2011 12:02 p.m., Anton Smith wrote:
Thanks Dylan and everybody else for the excellent feedback.

I think I have gained quite a few good ideas on a node to use for this purpose. If anybody has information on the performance (PPS scaling) on the mikrotiks that would be of interest.

If anybody has information on 2nd hand 7710s from ALU also that would be interesting (and for that matter, cisco 72xx, juniper m5/7x, or the newer boxes as Dylan mentioned like the ME3600, including any required license/sw costs).

Thanks again,
Anton

On 21 December 2011 22:05, Dylan Hall <dylan@citylink.co.nz> wrote:
I've not read the details of the UFB handover, but assuming it's similar to HSNS and EUBA we have two ways of handling them.

On one hand-over we're using a Linux box. We're using the bridging code to combine one or more Chorus vlans into a bridge which we then put on a different interface out to our main network. We've got this going with single stacked (HSNS) and double stacked (EUBA) services.

On our other hand-over we're using a Cisco ME3600. We're using a service-instance per customer to map them into a bridge-domain. We've only tested this with single stacked services but I believe it should work with double stacked.

In both cases we're not doing any fancy queuing, instead relying on Chorus to do the rate-limiting for us. Also, we're interested in providing an end-to-end layer 2 service only which may make our requirements somewhat different than the average service provider :)

Happy to provide config snippets for any of the above if it helps.

Dylan



On Wed, 2011-12-21 at 16:12 +0000, Anton Smith wrote:
Hi all,


Given the handover options from Chorus for UFB (http://www.chorus.co.nz/file/1591), does anybody have any recommendations about a cost effective GE based node that can do QinQ happily? Port density probably only needs to be 2-4 ports. Control/forwarding redundancy  not required.


And, preferably with the capability to set individual policers and/or queue rates on each sub-interface.


I don't know if chorus give you the option to do the rate limiting for you or not - does anybody know?


Cheers,
Anton
_______________________________________________
NZNOG mailing list
NZNOG@list.waikato.ac.nz
http://list.waikato.ac.nz/mailman/listinfo/nznog




_______________________________________________
NZNOG mailing list
NZNOG@list.waikato.ac.nz
http://list.waikato.ac.nz/mailman/listinfo/nznog