HP StorageWorks Enterprise Virtual Array Cluster FIO Starter Kit Betriebsanweisung PDF herunterladen (Seite 120)

SAN switches

All switches on a fabric must be from the same vendor. It is permissible for one fabric to contain

switches from one vendor and the other fabric to contain switches from a different vendor. Switches

are not supported in vendor neutral roles (or interoperability mode).

High bandwidth devices (such as tape backup servers and storage arrays) often use the same SAN

switches as the EVA Cluster components. Because the VSMs also perform data movement tasks, they

should be considered a high bandwidth device.

Fibre Channel links

Congestion occurring in FC is likely to cause problems because the protocol does not provide effective

mechanisms for relieving the congestion. Problems on a congested link or fabric can range anywhere

from slow response times, to discarded I/O, to loss of access to the fabric. Any link has the potential

for “fan-in,” where multiple links funnel into a single link, or a smaller number of links can become

overloaded. The most well known examples are ISLs (inter-switch links).

Take peak loads into account and add some margin above and beyond that number. It cannot be

reiterated enough that once a FC link becomes saturated or congested, major changes may be required

to get it out of that state. If ISLs are being used, it is best practice to set up alert levels in the switches

to ensure that notification of problems are realized.

Limit the number of switch hops from the servers to the DPMs, and from the DPMs to the storage, to

a maximum of three hops. In no case should the server-to-storage route exceed a total of seven switch

hops.

Mixing SAN-level virtualization with non-virtualized environments

Environments in which some logical units (LUs) are accessed directly from the array and other LUs are

accessed by the DPMs are supported. The same back-end LU must not be presented to the EVA Cluster

and directly to servers, or data corruption will occur. Naming conventions that help distinguish between

these two presentations is one way to make it easier to avoid and troubleshoot this kind of problem.

Setup volume configuration

Field experience has shown that many issues arise when access to the EVA Cluster setup volumes is

slow. The VSM event log must be monitored for messages indicating slow setup volume updates.

If the event log indicates a recurring setup volume problem over several hours the following actions

should be taken:

• Verify that the setup volumes are made from similar performance-based storage.

• Verify that the volumes are made from high performance RAID1 storage.

• Verify that the arrays containing the setup volumes are not extremely loaded.

• Move the setup volumes to less busy or faster arrays.

• Move the setup volumes to their own disk group or volume group.

• Move some of the setup volumes to dedicated arrays.

• Reduce the number of setup volumes.

Configurations with large numbers of service-enabled volumes (for example, thin provisioned or with

PiTs) generate the most demand for setup volume access.

Configuration best practices120

1 2 ... 115 116 117 118 119 120 121 122 123 124 125 ... 207 208

Keine Kommentare

HP StorageWorks Enterprise Virtual Array Cluster FIO Starter Kit Betriebsanweisung Seite 120