Database Reference
In-Depth Information
Scaling read throughputs
Unlike the write operation where only the master node is eligible for writing, reads can be
performed locally and can be served by any node in the cluster. Read operations in HA
Neo4j cluster increase linearly where any node in a cluster can serve the read request and
new nodes can be added at any point of time.
Let's discuss strategies, which if implemented, can provide desirable throughputs for read
operations.
Load balancer
Neo4j HA cluster does not provide any load balancing that can distribute the load over the
nodes in a cluster, but we can introduce a software or hardware load balancer for distribut-
ing the load equally over the various nodes in the cluster.
For example, we can introduce a software load balancer such as HA proxy—
ht-
tp://www.haproxy.org/
or Apache Proxy—
http://httpd.apache.org/docs/2.2/mod/
mod_proxy.html
,
which would intercept REST calls, and based on routing rules, can deleg-
ate the request to the appropriate node in a cluster.
