A white paper by John Vossler, a UNIX and Linux infrastructure consultant, presents a case study of designing two sites for a cluster to prevent loss of function of the cluster in case of a disaster. The paper outlines selecting a co-location hosting company with sites in different cities, setting up the Ethernet and SAN network connections, and how the Sun hardware and Cisco switches were set up and connected.
Clusters are designed to eliminate a Single Point of Failure (SPOF). By distributing hardware in noncontiguous locations in a data center, a SPOF can be eliminated as can the need for Disaster Recovery (DR) and Business Continuity planning. Since these plans can be problematic to implement in the event of a failure, it would be better not to have to rely on the DR plan by having part of your HA cluster outside the disaster footprint.
The paper presents an illustration of the stretched cluster topology that shows the logical connectivity of each site's major components. Vossler tested the configuration to make sure that if Internet connection was lost from one of the two sites, it would not impact the applications or user access to them. The goal of only a short failover interval outage was achieved for this phase of the project with the next goal to eliminate the failover outage.
Challenges for this phase of the implementation project were latency and VLAN and SAN stretching. Improvements to the project are also proposed in the white paper. In conclusion, Vossler said, "WAN based clusters are by definition disaster tolerant and eliminate the need for disaster recovery plans and business continuance plans." Vossler asserts that "infrequently exercised procedures are the most prone to failure, thus fail at the time they are most needed".
Get full details about the project by visiting the IT Infrastructures web site for a link to request the full white paper to be delivered via email.
Read More ...