Hi, I'm looking at setting up a haproxy cluster, which is expected to hit 300.000+ active simultaneous connections.
Normally haproxy sends requests to the backend, having exchanged the source ip, with it's own. As far as I can gather - that would mean it will run out of ports, when it hits about 64k connections being open. But perhaps ports can be reused, if they are used against different backends ? (ie. the max open connection amount in the linux tcp stack becomes 64k * backends-available) I've googled for xmpp and haproxy, and found a few examples, but noone talking about what happens at that size.. Also - I need one client ip - to stay on the same backendserver (while it has the connection open to the frontend) - so the connection should be "backend sticky" for the duration of the client connection.. My other idea was to do "direct routing" - as Linux Virtual Server calls it - where the loadbalancer simply forwards the packages using MAC address, and the source ip stays the clients ip.. but I'd like to use haproxy - as I have most experience with that. Can anyone shed some light on how that part works? Thank you in advance -- Regards, Klavs Klavsen, GSEC - [email protected] - http://www.vsen.dk - Tlf. 61281200 "Those who do not understand Unix are condemned to reinvent it, poorly." --Henry Spencer
