Specifications

DATA CENTER BEST PRACTICES
SAN Design and Best Practices 61 of 84
STORAGE
Storage arrays have evolved signicantly over the last few years. Performance has increased, capacities have
exploded, and more LUNs are supported than ever before. The performance and capacity of low-end arrays have
also improved. New features include the following:
•Some arrays time out and reset their ports if they do not receive acknowledgements from the connected host
after specic intervals.
•New behaviors include using in-band Fibre Channel for control purposes, which can put extra stress on FC port
buffer usage.
Note that storage array performance can degrade over time, which can be attributed to factors such as these:
•Miscongured LUNs can impact performance.
•Provisioning strategies can favor capacity over usage. An example of this might be a policy that dictates
the number of terabytes allocated per storage port. Applications accessing the LUNs can overload the array
capacity in order to service the requests.
Fixing degraded array performance is never easy. It usually involves some data migration or array reconguration.
Bottleneck Detection can be used to detect these conditions early, and changes can be implemented before
performance degradation becomes chronic.
Design Guidelines
•Be careful if you deploy mixed arrays with different performance characteristics. Experience has shown that it
is very easy for a Tier 3 storage array, depending on how it is used, to impact the performance of arrays in the
same fabric. Troubleshooting in these situations is very difcult.
•Control the number of LUNs behind each storage port based on the type of usage they will receive.
•Check on any special short-frame trafc to avoid frame congestion at array ports. It may be necessary to
increase the number of buffers at the array port to accommodate the extra control trafc.
•Use advance Brocade FOS threshold timers to monitor hosts and storage arrays to ensure that array ports do
not reset due to a high-latency host, and thus do not adversely impact other connected hosts.
Monitoring
•Bottleneck Detection is indispensable; many high-latency array ports can be identied and their performance
problems addressed before issues come to the attention of the server administrator.
•Use Brocade Fabric Watch to monitor Class 3 frame discards due to TX timeout so that severe latencies on
storage array ports can be identied.
Storage Virtualization
Storage virtualization enables LUNs accessed by servers to be abstracted from the physical storage (typically
storage arrays) on which they actually reside. (These are not the same as traditional storage array LUN
allocations, which can also be viewed as a form of virtualization.) Virtualized LUNs that are disassociated from
their actual storage allow for more exible storage provisioning processes. Performance may also improve, as
the virtual LUNs can be striped across multiple storage arrays.
There are two general types of storage virtualization: one uses an external controller (called in-line virtualization),
and in the other, the virtualization occurs inside a storage array. In-line solutions are slightly more exible,
because they can use physical storage from a variety of sources and vendors.