SAN Configurations

  • Article

Applies To: Windows Server 2003, Windows Server 2003 R2, Windows Server 2003 with SP1, Windows Server 2003 with SP2

Your SAN configuration directly affects the speed and reliability with which your systems can access storage devices on the SAN. Two critical considerations are your choice of a Fibre Channel SAN topology and the use of redundant components in the access path to storage devices.

Supported Fibre Channel Topologies

Windows Server 2003 supports all Fibre Channel SAN topologies (point-to-point, loop, and fabric). For attaching single storage devices to Windows Server 2003 systems using a Fibre Channel host bus adapter (HBA), the point-to-point topology is ideal. There are, however, some limitations with respect to the loop (FC-AL) topology. Loop topologies are inherently limited in operation and scalability, because they often require full loop reinitialization when devices are added or removed from the SAN. This can cause data to be inaccessible for periods ranging up to minutes. Also, loops can only scale to support up to 127 devices, and, like token ring networks, only allow a single device on the SAN to transmit data at one time, which might not make full use of the available bandwidth on the SAN.

Windows Server 2003 supports FC-AL topologies when a single server is connected to storage. However, the only multiserver configuration supported on a loop is a single two-node cluster. This limitation also applies to the derivatives of FC-AL topologies: Fibre Channel Private Loop SCSI Direct Attachment (FC-PLDA), Fibre Channel Fabric Loop Attachment (FC-FLA), and Fibre Channel Private Loop Attachment (FC-PLA). This means that your SAN design should incorporate a true fabric, or switch-based fabric topology, using a fabric SAN switch (hubs or loop switches should not be used) for any SAN implementation that contains more than two components (servers or storage devices).

Important

  • Placing new devices on an FC-AL–based SAN causes Loop Initialization Primitives (LIPs), which appear to the operating system as storage bus resets. LIPs can result in data unavailability and even data loss, if a LIP interrupts a write operation that is only partially committed to the storage device. LIPs can also disrupt backup operations to tape devices and can cause your devices to be assigned new addresses. Therefore, you should use persistent binding. You can limit the propagation of LIPs by using only fabric-based SAN topologies and compatible devices.

For more information about supported Fibre Channel topologies, see article 317162, "Supported Fibre Channel Configurations." To find this article, see the Microsoft Knowledge Base link on the Web Resources page at https://www.microsoft.com/windows/reskits/webresources.

Redundant Components

Although storage data on a SAN offers many benefits, it also introduces additional components whose failure can result in loss of access to data. As on a LAN, the failure of an adapter, switch, or cable can interrupt or remove access to a storage device. The best means to provide for highly available storage is to implement a SAN that uses redundant storage (such as hardware RAID), redundant data access paths (including cables and switches) and server HBAs, and MPIO storage drivers.