5

Using distributed Erlang creates a mesh network between all the nodes, and that doesn't scale well to hundreds of nodes. How does riak-core deal with that?

Does it not use distributed Erlang (set -connect_all=false and use custom protocol for communication between riak nodes)? From a cursory look at the code, it doesn't seem like they are doing that.

Harish
  • 473
  • 2
  • 13

1 Answers1

4

I believe they recommend to limit the cluster to 100 nodes, I can't find the reference now, tho.

In clusters over 100 nodes, ping noise will be significant part of network traffic. Even bigger deployments require deep changes in Erlang VM and OS.

Here is the link to cluster capacity planning from their wiki: http://wiki.basho.com/Cluster-Capacity-Planning.html

user425720
  • 3,578
  • 1
  • 21
  • 23
  • The capacity planning page does not mention the upper limit on the cluster size. But they do say a moderately sized riak cluster is around 18 nodes. So I'm guessing anything beyond 100 nodes hasn't been attempted. – Harish Sep 25 '12 at 21:12
  • oh no, there is even blog post from basho employee with the script to deploy 100 nodes on EC2. But one thing to deploy, the other thing to use it. http://reiddraper.com/100-node-riak-cluster/ – user425720 Sep 25 '12 at 21:47
  • Erlang clusters seem to be good to 125-175 nodes, though there's research going into making them *far* more scalable http://www.release-project.eu/ – Sam Elliott Sep 27 '12 at 18:26