node connectivity and security

Justin Sheehy justin at basho.com
Sat Feb 13 00:24:51 EST 2010


On Fri, Feb 12, 2010 at 1:21 PM, wde <wde at free.fr> wrote:

> If I am right, the riak network layer is based on the erlang clustering system.

You are right, though not all communication between nodes is in Erlang
messaging (some is raw TCP and/or protobuffs) almost all of the
typical-behavior networking is in standard distributed Erlang.

The rest of your comments about both network setup and security are
spot-on as well.

> In WAN environment, I think that the real problem (as always) is the latency introduced by the network. I have no experience with riak in this context.

Completely correct, and we do not recommend running a single Riak
cluster across general purpose long-haul internet links.  The internal
protocols and models that Riak is built on for intra-cluster
communication are not intended for such highly-variable latency as
occurs in that situation.  Generally a single Riak cluster should be
composed of nodes with relatively local network connectivity between
them.

Our Enterprise product has long-haul cluster-to-cluster
multidirectional replication using entirely different communication
models for inter-cluster communication.  Just as failure conditions
are different across the internet than within the datacenter, so must
be our ways of dealing with them.

-Justin




More information about the riak-users mailing list