How did WhatsApp achieve 2 million connections per server?

If you have enough RAM it's not too hard to handle 1M or more connections on linux. These guys handled 10 million connections with a java application on a single box using regular CentOS kernel with a few sysctl tweaks:

sysctl -w fs.file-max=12000500sysctl -w fs.nr_open=20000500ulimit -n 20000000sysctl -w net.ipv4.tcp_mem='10000000 10000000 10000000'sysctl -w net.ipv4.tcp_rmem='1024 4096 16384'sysctl -w net.ipv4.tcp_wmem='1024 4096 16384'sysctl -w net.core.rmem_max=16384sysctl -w net.core.wmem_max=16384

Also they balanced /proc/irq/'s of the network adapter and added a tweak for better JVM work with huge pages:

sysctl -w vm.nr_hugepages=30720

Having two 6-core CPUs 57% loaded they served 1Gbps over 12M connections in 2013.

But you need HUGE amount of RAM for that. The above test was on a server with 96GB RAM, and 36GB of those were used by the kernel for buffers of 12M sockets.

To serve 1M connections with similar settings you'll need a server with at least 8GB RAM and 3-4GB of them would be used just for socket buffers.

Note there are three things here:

1. Getting the server to support two million connections. This is usually just tweaking the kernel such that the number of simultaneous connections are allowed and such that the context associated with each connection fit in (wired) main memory. The latter means you can't have a megabyte buffer space allocated for each connection.

2. Doing something with each connection. You map them into userspace in a process (Erlang in this case). Now, if each connection allocates too much data at the user space level we are back to square one. We can't do it.

3. Getting multiple cores to do something with the connections. This is necessary due to the sheer amount of work to be done. It is also the point where you want to avoid locking too much and so on.

You seem to be focused on 1. alone.