First let's start by describing the current architecture:
All my servers is are connected for my own need to an internal 100G network using a CRS504-4XQ-IN. The 2x100G links are each split in 4x25G links. Internet is provided using a CCR2004-1G-12S+2XS via 2 FTTH link , 1 ISP is providing direct internet connection and a BGP session via PPPOE, the other via a GRE interface. I am for now only doing BGP failover between the interfaces. I failed to find a way for now to load balance between the 2 interfaces with IPv6. BGP allows me to retrieve a /48 IPv6 and a /28 IPv4.
The CCR2004 is connected to the CRS504-4XQ-IN using 1 25G link. Servers manage a BGP session to it to get internet. It is also providing internet via a 10G link to the LAN via a firewall. See the diagram below:
Here are the possible ways I'm thinking to optimise the cabling and network,, but I don't know which is enough efficient:
- I'm thinking to first replace the 25G link between the CCR2004 and the CRS504, which is done using a cable splitted in 4 links with only 1 used, by 4x10G LACP using DAC. WIll the performance of it OK with a CCR2004? Quid of the CPU usage?
- Is there a way using 1 CCR2004 to load balance Ipv6 and IPv4 efficiently between the 2 ISPs? Or better is there a way without loosing too much performance to pass 1 /56 (IPv6) and one /32 (IPv4) via one one of the ISP? Someone suggested that it would only be feasible efficiently using 2 CCR2004. Is this right?
- Alternatively is there a way to have both the CCR2004 and the CRS504 replaced by one router without compromising the power usage? Maybe the CRS518-16XS-2XQ or the CCR2216-1G-12XS-2XQ? Can we plug 1G link to the SFP28 ports ? I doubt it is feasible.