HP StorageWorks SAN Virtualization Services Platform Best Practices Guide (5697-0935, May 2011)

Core-edge switch fabric topology
Core-edge switch topologies introduce ISLs. Best configuration practice is that the DPMs, VSM
servers, and array controllers are connected directly to the core switches as shown with red lines
in Figure 3 (page 12). Edge switches are used to fan out to a larger numbers of hosts as shown
with the dark lines in Figure 2 (page 11). This design limits the use of ISLs for the less concentrated
host traffic, and avoids running the more concentrated DPM-VSM server-array traffic across ISLs.
When using ISLs, they must be monitored to ensure they are not oversubscribed and under
provisioned, leading to dropped frames and congestion. Trunking multiple links is one option to
avoid this issue.
Figure 3 Core-edge switch fabric topology
When designing a core-edge topology, be careful not to allow the edge switches an additional
path to the storage. Figure 4 (page 13) illustrates how you may inadvertently create a condition
where I/O may be sent from a core switch to the storage through an edge switch instead of taking
a direct path. An I/O request from the server first passes through the edge switch, then to the DPM
via links P1 and P2. The DPM traffic would normally pass through the P3 link directly to the storage.
However, if that link is broken, and a path P4 exists to the storage, the I/O must again pass through
link P1. This causes a potential congestion on this link. It is better to allow the I/O to fail over to
the other fabric than to use the P4 link.
12 Fabric topology