A while ago I wrote about my VyOS config for Init7’s Fiber7-X product. Since then there has been a number of breaking changes, and a few additions that I would like to cover.
I will copy/paste a lot of the narrative from that post, and avoid a bit of the abstract conversation that went with it, so that this stands on its own.
If you have questions or comments, hit me up.
In the previous post we were talking about what VPP was. Here we explore a little why it matters.
It’s a fair question. Surely its not logical to invest so much time and effort into something that has been described numerous times as “janky”. One of my engineers even now says, “I understand why you want to do it, but I don’t agree that this is the right solution”. She’s right too btw.
Linux Routing is becoming a thing with me. I cant decide if the motivation is the extreme cost of dedicated hardware, or the knowledge that with a little effort you can make a free/cheap thing into a giant killer. David and Goliath is a fun story I guess.
VPP has been on my radar now for a few years. I have tried and failed a few times to get it into production typically on the internet edge of a datacentre in place of something expensive like a Cisco ASR or a Juniper MX.
Last year I wrote extensively about my experience with deploying VyOS to support my new uber fast internet connection.
I learned a lot in the process, and for this past year it has mostly worked fine. I am one of those people that can’t leave things alone however, and I was always tinkering with the setup. The VyOS box itself was happily communicating at 10G, but I would find the internal LAN would get choked up a lot and rarely hit 1G even with extreme threading (say 50-60 conns). I got grumpy about this for no reason and started to tinker. TL:DR - This was a mistake…
So it turns out that if you want metrics from VyOS, your two options are SNMP or Telegraf (towards InfluxDB).
SNMP is one of those things that has been around so long, you think its good, but really, its trash. Its a 1990s technology that is mostly singlethreaded and provides you very very fuzzy numbers. 5 min averages are not that useful in situations like today where clients plausibly have access to gigabit+ grades of connectivity. If you cant observe let alone react to a microburst, you’re pretty much blindfolded in a modern DC network setting.