We are looking some data about using a CCR as a PPPoE server.
Our company currently has 10000 customers distributed in multiple servers with x86 ROS v5.x
When any server raises the connected count above 900 clients the service begins to degrade with high delay and loss of throughput, even when using the PCQ trick and no simple queues. The total aggregated bandwidth tends to 400-500 mbps
Because of this incidents, we are searching for other alternatives to provide internet access to our customers.
Do you have any experience about the real PPPoE aggregation limit of a CCR?
we use x86 core system with i7 cpu, linux based with accel-pppd system
we have large problems with implementations pppoe contrentrator with mikrotik over 1000 concurent sessions on 1 board
on same system mikrotik can handle about 1200 concurent session above that begin degradation
we have in peak about 4000 concurent pppoe sessions and use 2 accel pppoe concentrator without big problems
we have testing all network on 1 contretator and he manage all conections , so we estimated limit on accel up to 6-10000 connection depending on traffic intensity
we use two for redundancy
only plus for mikrotik is larger compatibility with cpe’s from different manufacturers (there is some minor problems on accel with some clients) and much faster pppoe connection reassociations (when connection from cpe drop or cpe reboot) but that can be fine tuned in accel probably, we working on that.
we used simple queue on concentrator, and have problem over 1200 conn
We have over 30 different types of packages, queue three dont simplified our setup
We used to use “simple queue” too, and have problems with it over 700/800 pppoe active sessions.
When we migrate queues config to “queue tree” schema, cpu and memory load dramatically decreases, then we get a stable config with 2000 pppoe active sessions.
