How does 100G work with 4x25G lanes

[JanL]

Hello Community,

We are currently dealing with 100Gbit/s connections. I have heard that for 100G with 4x25G lanes a session can only be 25G in size, so one session can only use one of the four lanes. On the other hand I also read that the traffic is split into 64 bit portions and then distributed over the lanes parallel. Does anyone know a good source where i can find more about this?

Many thanks
Jan

[AdrianO · Accepted solution]

When you split by 4x25G you have four ports at 25G rate. If those 4 ports service 4 interfaces on the same host/server and you use lacp to aggregate them, one session will use only one port.

[FredrikB-NN2 · Accepted solution]

Just like 40 G links, 100 G links use four lanes. This enables us to break out separate 25 G (or 10 G in the 40 G case) connections. This is NOT to be confused with how a 100 G link works when split into four 25 G links. In 1 x 100 G mode, it certainly behaves as a single 100 G link, enabling you to have one single TCP, UDP or whatever session to occupy the entire 100 G link if you need. If you were to split the 100 G interface into 4 x 25 G and create a LAG of those, then every flow or session would use one of those LAG members, limiting it to 25 G.

An interface in 1 x 100 G mode will behave just like that, one single 100 G link. When it comes to the split in 64 bit portions, you're probably referring to the 64b/66b encoding used by 10 and 100 G Ethernet (1 G uses 8b/10b), but that is totally transparent to the user, so nothing you will ever notice.

[AdrianO · Accepted solution]

When you split by 4x25G you have four ports at 25G rate. If those 4 ports service 4 interfaces on the same host/server and you use lacp to aggregate them, one session will use only one port.