Tivoli Storage Manager for Windows: Administrator's Guide


SCSI Tape Failover

MSCS does not support the failover of tape devices. However, TSM can handle this type of a failover pattern with the correct set up. TSM uses a shared SCSI bus for the tape devices. Each node (two only) involved in the tape failover must contain an additional SCSI adapter card. The tape devices (library and drives) are connected to the shared bus. When failover occurs, the TSM server issues a SCSI bus reset during initialization. In a failover situation, the bus reset is expected to clear any SCSI bus reserves held on the tape devices. This allows the TSM server to acquire the devices after the failover.

Failover Setup

  1. Ensure that Windows Advanced Server or Datacenter Server is installed on all computers in the cluster.
  2. Ensure that your hardware configuration meets the following criteria:
  3. Ensure the SCSI bus is configured prior to installation.
  4. Ensure each device on the shared SCSI bus has a unique SCSI ID, both SCSI controllers and tape devices.
    Note:
    Most SCSI controllers default to SCSI ID 7. Thus one of the SCSI controllers must have its ID changed to a different number, such as SCSI ID 6. This ensures that the host adapters have the highest priority on the SCSI bus.
  5. Ensure that the SCSI bus resets are disabled. You will want to do this because there is a tendency for SCSI controllers to reset the SCSI bus when they initialize at boot time. The bus reset can interrupt any data transfers between the other node and tape devices on the shared SCSI bus.
  6. Ensure that only tape devices are connected to the shared bus.
    Note:
    CD-ROM drives or removable media devices should not be used on the shared bus.
  7. Ensure that the device names assigned by either the Windows device driver or the TSM device driver (ADSMSCSI) are the same on both machines. The name assigned by ADSMSCSI contains both the port and bus value. If the names do not match on both machines, the TSM Tape Failover support cannot be configured.

Terminating the Shared SCSI Bus

The following presents methods for teminating the shared SCSI bus. You must terminate the shared SCSI bus as part of the initial setup of SCSI tape failover. Also, the shared SCSI bus must be terminated before you bring a server back online.

There are several different methods that can be used to terminate the shared SCSI bus:

Note:
Any devices that are not at the end of the shared bus must have their internal termination disabled.

Configuration Considerations

Whether you will configure your system to include clusters depends on your business needs. It requires a great deal of planning. In addition to assuring the right type of hardware and the applicable software, the focus of clustering is the failover pattern. When a node fails or needs to be taken off-line, which node or nodes in the cluster will pick up the transaction processing? In a two-node cluster there is little planning necessary. In a more complex arrangement, you want to give consideration to how your transaction processing is best handled. A form of load balancing among your nodes needs to be accounted for so that you maintain peak performance. Another consideration is to ensure that your customers do not see any lag and little drop in productivity.

MSCS requires each TSM server instance to have a private set of disk resources. Although nodes can share disk resources, only one node can actively control a disk at a time.

Is one configuration better than the other? To determine your best installation, you need to look at the differences in performance and cost. Assume you have a TSM server-dedicated cluster whose nodes have comparable power. During failover, the performance of a configuration may degrade because one node must manage both virtual TSM server instances. If each node handles 100 clients in a normal operation, one node must handle 200 clients during a failure.

The planning steps you might use and the end result of such planning are covered in the Tivoli Storage Manager for Windows Quick Start. Suffice it to say that clustering takes planning to ensure the optimal performance of your system.


[ Top of Page | Previous Page | Next Page | Table of Contents | Index ]