On Wed, 9 Feb 2022 at 22:19, Tom Beecher <beecher@beecher.cc> wrote:
Side note, am I missing something obvious where I can’t just have hardware routers strip ICMP, pipe it separately, put 500 VMs behind 4 vLBs and let the world ping the brains out of it?
Seems like a lot of overhead for zero benefit.
I'm not going to opinion on the quantity of benefits, but this thought could lend a razor from Occam. NPU based boxes, like JNPR Trio, NOK FP, Huawei Solar, CSCO Lightspeed et.al. could easily respond to ICMP echo and TTL exceeded in NPU for microseconds of delay and nanoseconds of jitter at higher performance and lower cost compared to transing it, i.e. ping responder would become negative cost. Only reason they don't is because customers are not asking for it. Further, we could have a global anycast address, like we already have for 6to4 relays, where a well-known ping responder exists. And anyone who welcomes responding to pings, configures this address to all the device loopbacks which they want to include, advertise those loopbacks in IGP or iBGP and advertise the /24 aggregate in eBGP. -- ++ytti