Google needs at least one BGP session between the GGC Router and the ISP. Even if we support multihop, we prefer a BGP session with the directly connected neighbor. Alternatively, contact your GGC coordinator to configure a static route.
The maximum number of BGP sessions supported is 4.
GGC peers with private ASNs are not supported. Private ASN ranges are:
-
64512 - 65534
-
4200000000 - 4294967294
Google doesn’t support configuring a BGP session that’s kept shut and is only active when the main neighbor is down. ISPs can influence which session is to be used via prefix advertisements. BGP sessions that are down trigger alerts.
Prefix advertisements
Google expects to receive user prefixes on the BGP session.
Google will advertise node's subnets (both IPv4 and IPv6). The ISP needs to re-advertise the node’s subnet and the interconnects to the internet. Google should be able to reach these subnets from any routable IP.
BGP Multihop
BGP multihop is supported. The default TTL is 4. The BGP neighbor IP for multihop BGP sessions with the GGC Router is always the first IP in the GGC node subnet. We don’t support multihop BGP sessions on interconnect IPs. GGC Router will use IPs on your side of interconnects to reach multihop BGP peers.
BGP Multipath
BGP multipath is supported. An ISP can configure a maximum of 4 paths.
Filters
There should not be filters or transparent proxies placed in the path of communications between the GGC node and Google’s back-end servers, or between the GGC node and users. Please refer to the Access Control section for more details.
Supported BGP attributes
A node with interconnects to 2 (or more) ISP routers that advertise the same prefixes:
- Google will ECMP. If the ISP wants to TE/only use one of the links, please AS prepend on that link
- Communities are not supported for selecting a link on the same GGC router - communities only work for selecting the node itself. More details are available here
- MED (Multi-Exit Discriminator) is not supported.
- If the ISP has redundant interconnects, ideally they should be equal sized and the same prefixes received over both. This allows us to ECMP
A node with interconnects to 2 (or more) ISP routers that advertise different sets of prefixes:
- Prefixes will be served respecting the route nexthop, as received via BGP