Specifications
DATA CENTER BEST PRACTICES
SAN Design and Best Practices 54 of 84
Monitoring
•Bottleneck Detection becomes very important here. Use it to monitor latencies on both the hypervisor and
storage ports to identify high latencies as soon as you can. Address the latencies as soon as possible.
•Brocade Fabric Watch is essential in early notication of potential issues in the fabric. Given the much
higher concentration of I/O due to the server consolidation, you should closely monitor trafc levels. There
is a continuing integration between Brocade Fabric Watch, Advanced Performance Monitoring, Bottleneck
Detection, and Port Fencing that you should exploit to the fullest.
•Monitor the Class 3 frame discards (C3TX_TO) through Brocade Fabric Watch as well. They are a strong
indication of high-latency devices.
SCALABILITY AND PERFORMANCE
Brocade products are designed with scalability in mind, knowing that most installations will continue to expand
and that growth is supported with very few restrictions. However, you should follow the same basic principles
outlined in previous sections as the network grows. Evaluate the impact on topology, data ow, workload,
performance, and perhaps most importantly, redundancy and resiliency of the entire fabric any time one of the
following actions is performed:
•Adding or removing Initiators:
– Changes in workload
– Changes in provisioning
•Adding or removing storage:
– Changes in provisioning
•Adding or removing switches
•Adding or removing ISLs
•Virtualization (workload and storage) strategies and deployments
If these design best practices are followed when the network is deployed, then small incremental changes
should not adversely impact the availability and performance of the network. However, if changes are ongoing
and the fabric is not properly evaluated and updated, then performance and availability can be jeopardized.
Some key points to cover when looking at the current status of a production FC network are these:
Reviewing redundancy and resiliency:
•Are there at least two physically independent paths between each source and destination pair?
•Are there two redundant fabrics?
•Does each host connect to two different edge switches?
•Are edge switches connected to at least two different core switches?
•Are inter-switch connections composed of two trunks of at least two ISLs?
•Does each storage device connect to at least two different edge switches or separate port blades?
•Are storage ports provisioned such that every host has at least two ports through which it can access LUNs?
•Are redundant power supplies attached to different power sources?
•Are zoning and security policies congured to allow for patch/device failover?