On Thu, Aug 11, 2016 at 5:20 PM, Alia Atlas <[email protected]> wrote: > Jesse, > > It's a question of the benefit of the change versus the cost to modify the > hardware. > Changing one FPGA or ASIC or reprogramming a network processor is > significantly cheaper > than redesigning a system architecture. > > I certainly agree with and share the frustrations of seeing extensibility > built in and then abandoned > - over and over again. > In a large datacenter deployment we need extensibility to deal with changing conditions and threats. For example, maybe we need add some bits to the headers that are needed to address a newly discovered security vulnerability. The worse case scenario is that if we add those bits it pushes us beyond the limits of HW processing and extending the protocol is no longer feasible. If a vendor tells us that we just need to swap out a few hundred thousand parts to get a larger parsing buffer to move forward it would NOT be an acceptable answer.
We need to know upfront exactly what the capabilities of the hardware and the protocol are. If the hardware can't give us a reasonable guarantee of extensibility, then we'll simply avoid using those features and don't commit to a hardware solution. Extensibility is a requirement in the datacenter, hardware support is not-- at best the latter is just an optimization. Tom _______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
