I have a requirement to terminate a large number (thousands) of EoIP tunnels at a central location. I am thinking about using x86 blades to terminate the tunnels and RBs at the edge for the users.

Has anybody used EoIP on this scale?
Is there any reason I could not configure say 1000 or so EoIP tunnels on a fast x86 based router?
Are there any figures available for throughput on EoIP or shallI try it out and let you all know?

Thanks!

don't use EoIP for heavy traffic applications.
use MPLS/VPLS instead.

only use EoIP for qucik'n'dirty emergency fixing stuff.

Eoip adds at least 42 bytes to each packet. That adds up quick.

the overhead isn't the problem IMO.
cpu-usage is.

Routerboard products will silently drop or delay packets with
EoIP under stress, even if CPU-Usage shows less than 20%!
X86 products can handle EoIP better.

I tried to handle one TCP-stream through a EoIP-tunnel between a x86-router and a RB1200.
The TCP-Stream dropped to 5Mbps. Nevertheless I could achieve max datarate by stacking tcp-sessions.

In comparison a VPLS-tunnel simply handled even one tcp-session with full datarate...

I understand that EoIP is not.. Ideal.
But I need to transport Ethernet over IP and do it with fragmentation and resammembly of full size (1500 byte) frames inside the tunnel.

VPLS etc would be perhaps nicer, but then I need to use Ethernet over MPLS over (insert tunneling mechanism here) over IP, which sounds like even more overhead!

Tbe RB750 is the ideal CPE for this operation, its the core that causes issues. If only they supported L2TPv3, I could Ethernet over that and have something bigger in the core.

--
Leigh Porter

MPLS only adds little more confugration overhead. And has much less encapsulation-overhead, thus better performance.

You can use EoIP, for sure.
But be warned that high bandwidth might cause serious trouble.

Indeed I would love to use MPLS, but I only have an IP transport between sites available. So it MUST go over IP.

I completely agree that the whole thing is rather nasty