Cluster Interconnects

What is clustering and why is it interesting?

Clustering is a computer architecture to create a network of servers that cooperate with each other. This cooperation has two forms. The first, clustering for availability, allows a customer's application to keep running on a secondary server should the primary server fail. The second is clustering for scalability where the customer's application is parcelled out over two or more computer systems to gain performance.

What does it take to cluster?

Clustering of servers requires three things:

• A cluster-aware operating system. This is our Solaris Operating Environment.
• Clustering software as the interface between the Solaris Operating Environment and our customer's application. Customers have a choice here depending on the nature of the application: Sun Cluster or HPC ClusterTools.
• An interconnect between the servers for heartbeat messages and data and application coordination information.

The figure shows a typical two node high-end cluster with a Sun Fire 15K server clustered with a Sun Fire 6800 server.

We offer our customers a choice of cluster interconnects at various performance levels and price points. The goal is to match the performance of the interconnect to the requirements of the customer's application. Important things to consider are the data rate of the interconnect and the latency - the delay.

• Fast Ethernet. A basic inexpensive cluster interconnect suitable for cluster failover. Data rate in the 5 MBps range. Latency about 100 microseconds.

• Gigabit Ethernet. 10 times the data rate of Fast Ethernet (50 MBps) but the same latency of 100 microseconds.

• Scalable Coherent Interconnect (SCI). A specialized cluster interconnect for Sun Cluster. Data rate of 70 MBps with latency reduced to 10 microseconds.