[justinsaccount]

This doesn't really replace something like haproxy. In the diagrams, the "real" servers that they are load balancing to are L7 load balancers like HAProxy (using https://github.com/facebook/proxygen most likely)

The problem this solves is is when you are using haproxy but need redundancy or a single server is no longer capable of handling the load and you need to scale out to 2+ servers.

You could run 4 haproxy servers and use round robin DNS to load balance between them, but as they say, this has the problem of:

> compared to DNS it doesn't need to wait for TTL to redirect traffic from failed L7 lb.

Many people likely use anycast/ECMP for this case, but that has the other issue they mention:

> compared to anycast-based solution, L4 lbs are more resilient to the networking related issues (which triggers mass ECMP reshuffle)/ enable better handling of adding/removing of L7 lbs from the pool as well as provide better support for unequal load balancing.

[workinthehead]

Why does nobody in this thread seem to realize that HAProxy works just fine over layer 4?

[justinsaccount]

> The problem this solves is is when you are using haproxy but need redundancy or a single server is no longer capable of handling the load and you need to scale out to 2+ servers.

Please describe how HAProxy running in l4 mode is a solution to this problem.

[tehnerd]

haproxy terminates tcp session. ipvs/katran does not. the whole purpose of ipvs/katran is to scale out layer of load balancers, which terminates tcp (transport in general) sessions. so tl;dr ipvs/katran solving the problem of scaling out "haproxy/nginx/varnish/w/e else" layer