,
10 tweets,
5 min read
Read on Twitter
@burneeed Our work predates gNMI and gNMI doesn't solve anything - it's just different. We prefer NETCONF, which is a standard, over gNMI, which is not. From a technical point of view there is nothing really wrong with gRPC although it is lacking some of NETCONF features.
@burneeed If you are the Goog and run gRPC on N million computers then you want the same stack on your network gear. Makes sense, I get that. If you are Facebook you want Thrift and if you are Microsoft you want Bond. NIH seems to be strong with these companies ;)
@burneeed I don't see why I would adopt it though as we are already happily using a standardised protocol, NETCONF, with more features, that we need and that solves the problems we have.
@burneeed I'm not saying NETCONF is perfect, it certainly isn't and I would love to see improvements to it. With time I will probably push and work on some of those improvements myself if no one else will. gRPC doesn't do a better job though so there's no upside to switching for me.
@burneeed There are also simple preferences, like gNMI is gRPC which is HTTP. If you're a web shop (google.com) then you are used to HTTP and probably love it. NETCONF runs over SSH. We like the operational side of SSH as our entire ops org understand the security model of SSH
@burneeed I'm not saying HTTP is a better bottom layer transport than SSH nor the other way around - just that one might be more suitable for you due to various reasons.
@burneeed NETCONF supports three phase commits which I think is pretty useful for networks where you risk shooting yourself in the foot. Our NMS uses this so that transactions that cause network connectivity issues are automatically rolled back.
@burneeed Maybe I should write a longer blog post about this. I get this question rather frequently. Sorry if I sound defensive, I'm not or don't try to be :) so much hype about it when it really doesn't do anything new. Very happy to discuss the topic! :)
@burneeed We've been using native NETCONF on IOS XR for many years now. When I started at DT a bit more than four years ago I think we were on XR 5.1 with a special build for us that had NETCONF. Two years later the gNMI spec was written. It didn't solve anything.
@burneeed The problems you run into with NETCONF on XR has nothing to do with NETCONF, it's all with the underlying infrastructure that was never built for this type of programmatic access. JUNOS did it right two decades ago, thanks again @philatjuniper. XR didn't get it right :/
