Stretched system configuration by using interswitch links

You can use interswitch links (ISLs) in paths between nodes of the same I/O group. If the cable distance between the two production sites exceeds 100 km, potential performance impacts can result.

Using ISLs for node-to-node communication requires configuring two separate SANs, each of them with two separate redundant fabrics:
  1. Configure one SAN with two separate fabrics so that it is dedicated for node-to-node communication. This SAN is referred to as a private SAN. This private SAN can be used by more than one stretched system if it has enough bandwidth for all these systems. For more information, see Additional bandwidth requirements.
  2. Configure one SAN with two separate fabrics so that it is dedicated for host attachment and storage system attachment.

The network hardware that carries the ISLs must maintain the physical separation and independence of the two redundant fabrics. For example, do not connect the two fabrics into a single dark fibre link. If there are two dark fibre links, dedicate one link for each fabric. Do not cross-connect the two fabrics onto each of the two links.

Stretched system configuration rules for stretched system configurations that use ISLs

In a stretched system configuration, a site is defined as an independent failure domain. Different types of sites protect against different types of fault. For example, if configured properly, the system continues to operate after the loss of one failure domain.

However, the system does not guarantee that it can survive the failure of two sites.

  • For every storage system, create one zone that contains ports from every node and all storage system ports, unless otherwise stated by the zoning guidelines for that storage system. However, do not connect a storage system in one site directly to a switch fabric in the other site. Instead, connect each storage system only to switched fabrics in the local site. (In stretched system configurations with ISLs in node-to-node paths, these fabrics belong to the public SAN).

    For stretched systems that use the enhanced configuration functions, storage systems that are configured to one of the main sites (1 or 2) need be zoned only to be visible by the nodes in that site. Storage systems in site 3 or storage systems that have no site that is defined must be zoned to all nodes.

  • Each node must have direct Fibre Channel connections to at least two SAN fabrics. One is a public fabric and the other is private, in its location. For an example configuration, see Figure 1.
  • Some service actions require the ability to do actions on the front panel or through the technician port of all nodes in a system within a short-time window. If you use stretched systems, you are required to assist the support engineer and provide communication technology to coordinate these actions between the sites.
  • The storage system at the third site must support extended quorum disks. This information is available in the interoperability matrices that are available at the following support website: http://www.ibm.com/support.
Each SAN consists of at least one fabric that spans both production sites. At least one fabric of the public SAN includes also the quorum site. You can use different approaches to configure private and public SANs.
  • Use dedicated switches for each SAN.
  • Use separate virtual fabrics or virtual SANs for each SAN.
    Note: ISLs must not be shared between private and public virtual fabrics.

To implement private and public SANs with dedicated switches, any combination of supported switches can be used. For the list of supported switches and for supported switch partitioning and virtual fabric options, see the interoperability website: www.ibm.com/support.

Like for every managed disk, all nodes need access to the quorum disk by using the same storage system ports. If a storage system with active/passive controllers (such as IBM® DS3000, IBM DS4000, IBM DS5000, or IBM FAStT) is attached to a fabric, the storage system must be connected with both internal controllers to this fabric. This connection is illustrated in Figure 1.

Figure 1. Stretched system nodes with ISLs, DS3000, DS4000, and DS5000 connected to both fabrics
Stretched system nodes with ISLs, DS3000, DS4000, and DS5000 connected to both fabrics

By using FCIP, passive WDM, or active WDM for quorum site connectivity, you can add to the extension. The connections must be reliable. It is strictly required that the links from both production sites to the quorum site are independent and do not share any long-distance equipment. FCIP links are supported also for ISLs between the two production sites in public and private SANs. A private SAN and a public SAN can be routed across the same FCIP link. However, to ensure bandwidth to the private SAN (see also Additional bandwidth requirements), it is typically necessary to configure FCIP tunnels. Similarly, it is permissible to multiplex multiple ISL links across a DWDM link.

Note: It is not required to UPS-protect FCIP routers or active WDM devices that are used only for the node-to-quorum communication.

A stretched system configuration is supported only when the storage system that hosts the quorum disks supports extended quorum. Although the system can use other types of storage systems for providing quorum disks, access to these quorum disks is always through a single path.

For quorum disk configuration requirements, see the Guidance for Identifying and Changing Managed Disks Assigned as Quorum Disk Candidates technote at the following website: Guidance for Identifying and Changing Managed Disks Assigned as Quorum Disk Candidates.

Additional bandwidth requirements

The connection between each site in the system that is associated with the nodes must ensure a minimum bandwidth of 4 gigabits or two times the peek host-write I/O workload, whichever is higher. The exact usage of this bandwidth depends on whether the site for the system is configured for nodes and storage systems. If the system is not configured as site-aware, half of this bandwidth is required in the public SAN and the other half is required in the private SAN. If the system is configured as site-aware, all of this bandwidth is required in the private SAN. In any case, this bandwidth must be ensured for the traffic between nodes and also between nodes and external storage systems. This bandwidth that is dedicated for system traffic does not include the traffic between nodes and hosts. Hosts that access nodes in the other site require more intersite bandwidth.

Note: To avoid performance bottlenecks, each system node requires two ports that operate at full speed (for example, 2 x 8 gigabits) worth of bandwidth to other nodes within the same system.

For example, a stretched system (no matter how many nodes) with a total of 4 gigabits of stretched system bandwidth on the private SANs is a valid configuration when the peek host-write I/O workload does not exceed 200 megabytes per second.

Operating an eight-node system at its maximum performance requires 64 gigabits of stretched system bandwidth on the private SANs.