On 1/3/21 1:50 PM, Mark Delany wrote:
On 03Jan21, Brandon Martin allegedly wrote:
I was thinking more in the original context of this thread w.r.t. potential distribution of emergency alerts. That could, if semi-centralized, easily result in 100s of million connections to juggle across a single service just for the USA. While it presumably wouldn't be quite that centralized, it's a sizable problem to manage. Indeed. But how do you know the clients are still connected? And if they aren't, there is not much a server can do beyond discarding the state. Presumably the client would need to run a fairly frequent keep-a-live/reconnect strategy to ensure the connection is still functioning.
Which raises the question: how long a delay do you tolerate for an emergency alert? I think the end result is a lot of active connections and keep-a-live traffic. Not really quiescent at all. In the end, probably just as cheap to poll a CDN.
I just sent some mail to the myshakes folks at UCB asking if they have an achitecture/network document. In their case for earthquakes it need to be less than ~10 seconds so they are really pushing the limit. If they get back to me, I'll share it here. Mike