Question / Problem
Certain operations in a CQ4/CQ5 authoring cluster (e.g. replication) are considerably slower when triggered from a slave cluster-node.
Answer / Resolution
The latency is caused by delays on the TCP layer: per default, small data packets are concatenated into bigger frames before actually sending them across the network to prevent network congestion. This has a negative performance impact on the TCP communication between the slave and master cluster-nodes and can be improved dramatically by setting the
TCP_NODELAY flag on TCP connections. This causes every data packet to be transmitted immediately.
Attached to this article is CRX hotfix pack 220.127.116.11 which causes the TCP connections used for cluster-communication between master and slave cluster-nodes to be opened with the
TCP_NODELAY flag set.