Re: [hybi] NAT reset recovery? Was: Extensibility mechanisms?

[Jamie Lokier <[e-mail]>]

Pieter Hintjens wrote:
> > However for packet efficiency, synchronising timing by using
> > request-response, or another synchronisation method, may still be
> > advantageous.  Even though it's more complicated than async keepalive.
> 
> Complexity is always worth removing unless the overhead is significant
> and since KAs can be entirely switched off when there is traffic at
> all, and then reduced to once per 5 or 10 or 30 seconds, packet
> efficiency seems irrelevant here.

I'm talking about when there is no traffic.  Some applications are
idle nearly all the time; then KAs are by far the dominant cost.

Three packets exchanged every 30 seconds is a ~26GB per thousand
user-months.  You might transfer 1GB of real data in the same time.

If you have a million connections doing KAs all the time, you can
go bankrupt paying for KAs while the real data is affordable.
Same with lots of open tabs and WS apps on a mobile browser.

It's not worth obsessing over a few percent, but halving the cost is
sometimes worth it.  But anyway, leave that to each implementation.

Just make sure the protocol supports (a) ping requests, and (b)
request to set async KA idle time, and all will be fine.  [(b) is
as easy as negotiation and more versatile.]

-- Jamie