Deployed no but just a matter of time - I doubt there are huge latency benefits that you don’t already see from the current plethora of optimised DC switches and code that’s available and I’m totally unconvinced that low buffer solutions are right for anything outside of the DC with today’s OTT steaming needs mixed with a lot of the access technology in cable, DSL and FTTP.
Our requirement is only to run code we know we need to run- simplifies security risks and stupid bugs in things we don’t use and allows focused testing and telemetry in the end to end service. Sent from my iPhone > On 17 Feb 2018, at 10:04, James Bensley <[email protected]> wrote: > >> On 17 February 2018 at 09:43, Neil J. McRae <[email protected]> wrote: >> However, you might want to look at P4 capable platforms which are looking >> very good for focused thin OS networking. > > Are you using P4 at all Neil? > > I'd be very keen to hear from anyone that is, what their experience has been. > > I've downloaded the BMv2 target [1] and started playing around with > code that would run on x86 to test. However, when it came to real > hardware testing Barefoot Networks seem to have gone from one ASIC > which supports P4, from when I first looked into P4, to multiple ASICs > [2] but they don't seem to be in stock anywhere?! > > If anyone has some P4 hardware, even in the lab, I'd love to hear > about your experiences. I've only tested it inside a VM which is > really more just letter one learn the syntax and architecture. I > wanted to test the impact of simplifying the forwarding pipeline to > the bare minimum to reduce latency, advanced/dynamic port buffers, and > iOAM. What have you managed to achieve with it that you couldn't > before? > > Cheers, > James. > > [1] https://github.com/p4lang/behavioral-model > [2] https://barefootnetworks.com/products/brief-tofino/ >
