Specifications

DATA CENTER BEST PRACTICES
SAN Design and Best Practices 26 of 84
Typically, a storage port services many applications, and the loss of that storage can severely impact all the
applications connected to it. One of the advantages of a core-edge design is that it is very simple to isolate
servers from their storage and ensure that any action applied to host port for a given behavior can be very
different than the action applied to a storage port for the same behavior.
DESIGN GUIDELINES
•Transaction-based systems: Make sure that ISL/UltraScale ICLs traversed by these systems to access their
storage do not contain too many ows. The fan-in from the hosts/initiators should not exceed a ratio of 10 to
1. Also ensure that there is as little interference from other applications as possible, to ensure that latencies
and congestion from other sources do not affect the overall performance of the applications.
•I/O-intensive applications: Bandwidth is the typical constraint for these systems. Modern fabrics typically
provide more bandwidth than is needed, except for the most powerful hosts. Take care to ensure that these
systems do not interfere with other applications, particularly if they are run at specic times or if batch runs
are scheduled. When in doubt, add more paths (ISLs or trunks) through the fabric.Clusters: Clusters often
have behavioral side effects that must be considered. This is particularly true during storage provisioning. It
is possible, for example, for a cluster to inundate the fabric and storage arrays with LUN status queried and
other short frame requests. This behavior can cause frame congestion in the fabric and can stress the control
processors of the arrays. Make sure that you spread out the LUNs accessed by the hosts in the cluster
across as many arrays as possible.
•Congestion: Trafc congestion (total link capacity regularly consumed) is remedied by adding more links or
more members to a trunk. Frame congestion is typically addressed by dealing with the nodes causing
the congestion.
•Misbehaving devices: It has been stated earlier that there is very little that a fabric can do to mitigate the
effects of a badly behaving device, other than to remove it from the fabric. Brocade supports a Brocade FOS
capability called Port Fencing, which is designed to isolate rogue devices from the network. Port Fencing
works with Brocade Fabric Watch to disable a port when a specic threshold has been reached. Port Fencing,
in combination with Bottleneck Detection, can be used for detecting and isolating high-latency devices from
impacting the rest of the devices in the fabric.
•Initiator and targets: If possible, isolate host and storage ports on separate switches for much greater
control over the types of controls that you can apply to misbehaving and high-latency devices. The effect on
applications is typically much less severe if a host is disabled versus disabling a storage port, which may be
servicing ows from many servers.
Monitoring
•Brocade Network Advisor is a powerful proactive monitoring and management tool that offers customizable
health and performance dashboards to provide all critical information in a single screen. With Brocade
Network Advisor, you can manage your entire network infrastructure.
•Use Brocade Fabric Watch to monitor switch and director resource consumption, port utilization, and port
errors. Brocade Fabric Watch is also used to trigger Port Fencing.
•Advanced Performance Monitoring is an end-to-end monitoring tool that can help when you encounter
congestion, including frame congestion.
•Bottleneck Detection is very useful in detecting latencies in devices and across links. It can help clarify
whether high buffer credit zero counts are actually a problem. Once device latencies have been addressed, it
is often useful to apply other controls, such as Port Fencing, to improve the resiliency of the fabric by isolating
new misbehaving devices or future high latencies.
•Brocade SAN Health is a free utility that provides a lot of useful information to the storage or SAN
administrator. You can look at ISL fan-in rations, get Visio diagrams of fabrics, verify rmware levels on
switches, and nd a host of other valuable information.