On 2 Dec 2014, at 12:41 am, Robert Hass <robh...@gmail.com> wrote: >> I think vMX can forward data. > > vMX indeed will be full-featured router. But my questions was related to > move part of control-plane (basically whole BGP part of rpd) to external > server. Maybe OpenFlow somehow helps here ? How openflow take care of eBGP > to customers ? Session should be on router or on OpenFlow controller ? OF > v1.3 just has been implemented in JunOS 14.x releases for MX series.
This is completely doable without the need for Openflow. The vMX is delivered as two VMs - a vRE and vPFE and logically they communicate in the same way as the RE and PFE do in an MX today eg: an internal ethernet switch sends control-plane information down to the forwarding engine. Whether this happens on a single host machine across a vSwitch, or across a physical network is now completely irrelevant, normal constraints not withstanding (congestion/loss etc). It's not a big stretch to imagine being able to one day ditch the vPFE and take a bunch of "head-less" MXs (hardware-based) acting as remote FPCs to a pair of centralised vREs - like JCS/TXM but on more commoditised server hardware, or even a Q-Fabric with MXs as nodes and commodity directors/interconnect. Given that control-plane traffic is pretty minimal, and with all the cool FIB localisation knobs coming down in 14.x, the scaling/design implications of this get pretty interesting. Heck - it should be possible to one day take the much-maligned MX80 RE and replace it with a remote intel-based monster, and get 64-bit Junos in the mix. Emulating the backplane for inter-FPC traffic is going to be a barrel of fun though! Ben _______________________________________________ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp
