Specifications

www.ibm.com/support/techdocs 29

Summary of Best Practices for Storage Area Networks

of the safety factor for traffic expansion, implement a spare ISL or ISL trunk. The fabric

still needs to be able to avoid congestion if an ISL were to go down due to issues such

as switch line card failure.

As stated earlier, a SAN design can exceed the best practice guidelines for ISL

oversubscription ratios of 7:1 with a clear understanding of the traffic patterns.

Whenever the “standard” 7:1 oversubscription ration is exceeded, then it highly

recommended that fabric bandwidth threshold alerts be implemented. Any time an ISLs

exceeds 70%, fabric changes should be planned and implemented to spread out the

load further.

Consideration must be given to the bandwidth consequences of a complete fabric

outage. While this is a fairly rare event, insufficient bandwidth could turn a single-SAN

outage into a total access loss event. _ Take the bandwidth of the links into account. It

is very common to have ISLs run faster than host ports, which obviously reduces the

number of required ISLs.

7 Servers

Clusters are another means of achieving highly available SAN environments. Server

clusters shift the focus from the hardware of a single server over to the application

which must be cluster aware. One interesting implementation being more involves the

use of server clusters in which the individual cluster nodes are not located in the same

data center. While this approach addresses disaster recovery and business

continuance concerns in a cost effective manner, it increases complexity in terms of

management. In addition, the use of split clusters is somewhat distance limited and

may require a significant work effort with tuning to account for the distance between the

cluster nodes.

In general, host resources such as memory and processing time are used up by each

disk volume that is mapped to the host. For each extra path, additional memory may be

used and some portion of additional processing time is also required. The user may

control this effect by using fewer larger volumes rather than lots of small volumes.

However it may require tuning of queue depths and I/O buffers to support this efficiently.

If a host does not have tunable parameters such as Windows, then it does not benefit

from large volume sizes. Conversely, AIX will greatly benefit from larger volumes with a

smaller number of volumes and paths presented to it.