Requirements to consider when you use a backup solution with Windows server clusters (887017)


The information in this article applies to:

  • Microsoft Windows 2000 Advanced Server
  • Microsoft Windows Server 2003, Standard Edition

INTRODUCTION

Microsoft provides a backup software solution with the Windows NTBackup software. Most Enterprise customers choose to use more robust enterprise-oriented backup software and hardware solutions. However, in choosing that solution, customers who use Windows server clusters have to recognize special considerations that apply to these servers. This article describes these considerations.

MORE INFORMATION

Considerations for cluster backup software include the following:
  • Cluster backup software should be "cluster aware" because it uses cluster APIs and can gracefully deal with the failover of a disk resource during the backup process.
  • While we do not recommend open file managers for use on a cluster, if any are installed they have to gracefully handle a failover of a disk resource and correctly release all handles to the disk. Failure to release open handles may cause the disk to appear to the receiving node to have the dirty bit set and will trigger a chkdsk /f on that disk resource.
Other considerations include the following:
  • The hardware and storage area network (SAN) fabric must be architected in such a way that tape devices do not share the same small computer system interface (SCSI) bus or host bus adaptors (HBAs) as the shared storage disks. For Fibre Channel, some vendors try to interpret bus resets into other Link Services commands, assuming the bus reset is only used to clear reservations and is issued by clusters only. In fact, bus resets are used within the Windows storage stack for recovering from command timeouts. If the command timeout was on the tape drive, and the reset is translated to a different Link Services command, the tape drive is likely not to recover and will not be available for use unless its power was cycled or the system was rebooted.

    Note The power being cycled could lead to other failures because some backup applications cannot correctly process the PnP notifications that will result.

    We do not recommend such a configuration with SCSI miniport or a monolithic full-port driver.

    Some storage vendors may offer their own solutions for coexistence of disk and tape on a single HBA with either a SCSIport miniport driver or a monolithic full-port driver. Support of these types of solutions is the responsibility of the storage vendor and not Microsoft, and the storage vendor must be the first point of contact for any disk related issues in these configurations. Microsoft supports the use of disk and tape on the same HBA using Storport miniport drivers.

    Microsoft recommends use of Storport, a new port driver, which is available for use in Windows Server 2003 for the above mentioned configuration. In the Storport environment, Windows will use LUN and target resets to clear reservations and timeout problems. This prevents a reset of devices that are not involved in the error condition. Cluster configurations based on the Storport driver that share tape and disk on the same HBA or SCSI bus can be supported by Microsoft.
  • Software that installs filter drivers, such as antivirus software or multipathing software, must be "cluster aware" and must be able to gracefully let the disk resource to failover to other nodes.
  • Some backup solutions use what is known as "server-less" backups. These solutions cannot be used with a server cluster because of the cluster's persistent binding to the HBA. Server-less backups must be performed to a mirror set which is implemented in the hardware of the SAN and does not involve the cluster software. All backup solutions that are implemented outside the server cluster's fabric are the support responsibility of the backup vendor and not Microsoft.
For additional information about Windows server clusters, visit the following Microsoft Web sites:

http://www.microsoft.com/ntserver/support/faqs/clusterfaq_hardware.asp

http://www.microsoft.com/ntserver/support/faqs/clusterfaq_deploy.asp

REFERENCES

For additional information, click the following article number to view the article in the Microsoft Knowledge Base:

321939 CHKDSK may be initiated if there are open handles when a cluster disk fails over

248998 How to properly restore cluster information

309395 The Microsoft support policy for server clusters, the Hardware Compatibility List, and the Windows Server Catalog

250355 Antivirus software may cause problems with cluster services

Modification Type:MinorLast Reviewed:1/4/2006
Keywords:kbwhitepaper kb3rdparty kbClustering kbhowto kbinfo KB887017 kbAudITPRO
©2004 Microsoft Corporation. All rights reserved.