Planning Fabric Metrocluster Configurations - IBM N Series Hardware Manual

8.4.1 Planning Fabric MetroCluster configurations

When planning and sizing Fabric MetroCluster environments, keep in mind these
considerations:
Use FibreBridges in conjunction with SAS Shelves (EXN3000 & EXN3500).
Provide enough ports/loops to satisfy performance (plan additional adapters if
appropriate).
Storage must be symmetric (for example, same storage on both sides). For storage that is
not symmetric, but is similar, file a RPQ/SCORE.
Keep in mind that N series native disk shelf disk drives are not supported with
MetroClusters.
Four Brocade/IBM B-Type Fibre Channel Switches are needed. For more information
about supported Switches and firmware in Fabric MetroCluster environments, see the
Interoperability Matrix at:
http://www-304.ibm.com/support/docview.wss?uid=ssg1S7003897
One pair of FC switches is required at each location. The switches need to be dedicated
for the MetroCluster environment, and cannot be shared with other systems. Remember
that you might need the following licenses for the Fibre Channel switches:
– Extended distance license (if over 10km)
– Full-fabric license
– Ports-on-Demand (POD) licenses (for additional ports)
Infrastructure / connectivity has these options:
– Dark fiber: Direct connections by using long-wave Small Form-factor Pluggable
transceiver (SFPs) can be provided by customer. No standard offering is available for
these SFPs for large distances (> 30km).
– Leased metro-wide transport services from a service provider: Typically provisioned by
dense wavelength division multiplexer/time division multiplexer/optical add drop
multiplexer (DWDM/TDM/OADM) devices. Make sure that the device is supported by
fabric switch vendor (IBM/Brocade).
– Dedicated bandwidth between sites (mandatory): One inter-switch link (ISL) per fabric,
or two ISLs if using the traffic isolation (TI) feature and appropriate zoning. Do not use
ISL trunking, because it is not supported,
Take care in designing fabric MetroCluster infrastructure. Check ISL requirements and
keep in mind that cluster interconnect needs good planning and performance.
Latency considerations: A dedicated fiber link has a round-trip time (RTT) of
approximately 1 ms for every 100km (~ 60 miles). Additional nonsignificant latency might
be introduced by devices (for example, multiplexers) en route. Generally speaking, as
distance between sites increases (assuming 100 km = 1 ms link latency):
– Storage response time increases by the link latency. If storage has a response time of
1.5 ms for local access, the response time increases by 1 ms to 2.5 ms over 100km.
– Applications, in contrast, respond differently to the increase in storage response time.
Some application response time increases by greater than the link latency. For
example, application A response time with local storage access is 5 ms and over
100km is 6 ms. Application B response time with local storage access is 5 ms, but over
100km is 10ms.
Take care in planning N6210 MetroCluster configurations, because the N6210 has only
two Fibre Channel initiator onboard ports and two PCI expansion slots. Because you use
one slot for the FC/VI adapter, you have only one remaining slot for a Fibre Channel
Chapter 8. MetroCluster
109