On 1 September 2016 at 03:02, Dragan Jovicic <[email protected]> wrote: > Things to note. Card has 4 PFE; each MQ chip is good for ~70Gbps give or > take +/- 10Gbps depending on packet sizes.
Correct. There are also some other complexities - in SCB+MPC2, you have 2xFAB capacity of what you need, this is reserved for binary multicast replication - in SCB+16XGE, all FAB capacity is spent for ucast, and multicast is replicated unary manner, hop-by-hop through each LC - If you have DPCE in same platform in default configuration you cannot get linerate from DPCE => MQ. This is because by default MQ will use all fabrics, where as DPCE won't. Im MX960 this means DPCE is sending 40Gbps through two fabrics, but MQ expects to receive it's share 40Gbps through three farbrics. So MQ will only give grants to DPCE at 40/3*2 = 26Gbps. This can be addressed by changing the fabric mode from default performance to 'redundant', causing MQ to use only two fabrics, allowing DPCE to send full 40Gbps to any given single MQ. -- ++ytti _______________________________________________ juniper-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/juniper-nsp
