Consider in Figure 2.32 a server-client cluster with an active-takeover configuration between two identical servers. The servers share a disk via a SCSI bus. The clients (PCs or workstations) and the Ethernet are fail-free. When a server fails, its workload is switched to the surviving server.

a. Assume that each server has an MTTF of 200 days and an MTTR of five days. The disk has an MTTF of 800 days and an MTTR of 20 days. In addition, each server is shut down for maintenance for one day every week, during which time that server is considered unavailable. Only one server is shut down for maintenance at a time. The failure rates cover both natural failures and scheduled maintenance. The SCSI bus has a failure rate of 2 percent. The servers and the disk fail independently. The disk and SCSI bus have no scheduled shutdown. The client machine will never fail.

1. The servers are considered available if at least one server is available. What is the combined availability of the two servers?

2. In normal operation, the cluster must have the SCSI bus, the disk, and at least one server available simultaneously. What are the possible single points of failure in this cluster?

b. The cluster is considered unacceptable if both servers fail at the same time. Furthermore, the cluster is declared unavailable when either the SCSI bus or the disk is down. Based on the aforementioned conditions, what is the system availability of the entire cluster?

c. Under the aforementioned failure and maintenance conditions, propose an improved architecture to eliminate all single points of failure identified in Part (a).

Client Client Client ... Ethernet Server 1 Server 2 SCSI bus Disk array FIGURE 2.32 An HA cluster with redundant hardware components.