Re: ROV Deployment (was LDPv6 Census Check)
On Sun, 14 Jun 2020 18:09:24 +0000 Randy Bush <randy@psg.com> wrote:
thanks to a few vendor engineers who implemented as skunkworks, to jay, you, and other large ops who have deployed, and to job who has taken over waving the pom poms, i am rather optimistic.
I concur. I asked our four major networks we (depaul.edu) had connectivity from if they supported ROV in early 2019. Cogent, HE.net, Internet2, and SeverCentral all said they do not and none had plans to deploy. Now two of those are doing some ROV. We've since replaced Zayo for Cogent and I asked them about it in late 2019. I know at least one of the remaining three is seriously looking at supporting it soon. Lots of reasons for the momentum, but two very noticeable reasons from my view were when AT&T announced their deployment and when Google got people worried about dropping routes. John
On Jun 16, 2020, at 7:53 AM, John Kristoff <jtk@depaul.edu> wrote: when Google got people worried about dropping routes.
That may have an impact down the road, but I doubt that really had that much impact on current deployments. -dorian
when Google got people worried about dropping routes. That may have an impact down the road, but I doubt that really had that much impact on current deployments.
i suspect different folk moved for various reasons. i appreciate the motion. while things are moving, the problem is that technical deployment sucks badly, from the arin disaster to broken relying party software to broken router implementations; i.e. every step in the chain. the only reason the mess is not blatantly visible is the fail soft design, aka notFound. the problem with fail soft is that you think you are protected when you are not. my inner naggumite is starting to wonder if fail soft was a mistake. randy
On Tue, Jun 16, 2020 at 11:51 AM Randy Bush <randy@psg.com> wrote:
router implementations; i.e. every step in the chain. the only reason the mess is not blatantly visible is the fail soft design, aka notFound. the problem with fail soft is that you think you are protected when you are not.
I don't see how we would have reasonably found these problems without large scale actually operating deployments. To me this seems like: ipv6 rollouts dnssec rollouts any other large system change we expected things to work like X, in reality they work a little differently AND we have software / systems problems which SEEM like non-problems (or even features!) which under stress/scale prove to be complications to be filed down.
my inner naggumite is starting to wonder if fail soft was a mistake.
would be hard to argue: "Sure! you should deploy, worse case when things go wrong in your deployment (which happens, always) you fall off the net!" fail soft at least for a while is ok... and helps get systems/people/scale.
participants (4)
-
Christopher Morrow -
Dorian Kim -
John Kristoff -
Randy Bush