Handling overload better

[dtaht]

Really great to see this new project. It is my hope that along the way, building a daemon that handles overload better both in cpu and in the protocol itself, emerges. Basically babel has a 4s tick, and when various problems accumulate that require processing that takes longer than that, bad things happen. This "feature" is common across many of our routing protocols today, not just babel, when there is a lot of churn in the network, which is precisely when you NEED a routing protocol to sort things out, things get worse than they should.

At a core level, I recommend using fdtimers to schedule certain transmissions, and to observe when that deadline is exceeded, and to do smarter things to shed load when it happens. Even the ancient top utility in procps tools does this correctly, why not a routing protocol? An underused capability in the babel protocol is the ability to set route announcements to have a longer or shorter expiry interval. Other means of coping with packet loss might include randomizing the start point of an announcement series, and using paced transfers.

I wrote a tool, years ago, that easily creates churn demonstrates other problems, here:

https://github.com/dtaht/rtod