notes blog about

Load balancer types

DNS Round Robin * simple to set up * difficult to control, not very responsive * if one replica dies, the clients will try to access it until the cache expires * no control over which backend (replica) receives the traffic

L3 and L4 * simple and fast * each TCP session is redirected to one of the replicas * L3 - all traffic from a given source IP will be sent to the same server regardless of the number of TCP sessions it has generated * L4 - track also source and destination port (finer granularity)

L7 * can examine what’s inside the HTTP protocol - like headers (cookies), URLs - and make decisions based on that * X-Forwarded-For: - header that can be inserted by a LB containing list of IPs traversed before the packet got to the LB

Load balancing methods

Solutions for “shared state problem” (ex. authenticated user)

User identity

Scale vs. Resiliency

Source: PoCSA