ServerNet Cluster 6780 Planning and Installation Guide (G06.29+, H06.06+, J06.03+)

Using the Long-Distance Option
ServerNet Cluster 6780 Planning and Installation Guide—527301-008
G-6
Recommended SPR for Long-Distance Cluster with
HP Integrity NonStop NS-series Nodes
Recommended SPR for Long-Distance Cluster
with HP Integrity NonStop NS-series Nodes
For long-distance clusters (with either two or three zones) that include Integrity
NonStop NS-series nodes in addition to S-series nodes, HP strongly recommends
T9050G09^AQB (released at G06.28). T9050AQB improves interprocessor
communication throughput over long-distance links between S- and NS-series nodes.
If you running H-series software, see the ServerNet Cluster Supplement for NonStop
NS-Series Servers for the recommended H-series SPR.
Restrictions
The numeric selector settings shown in Table G-3 are optimized for zone-to-zone links
between 5 and 15 km. Configuring a ServerNet cluster with the required long-distance
option software and the numeric selector settings shown in Table G-3, but with actual
lengths outside this range can degrade cluster performance and cause severe errors
that result in a loss of ServerNet connectivity between nodes.
Bandwidth Considerations
With the long-distance option, overall bandwidth is reduced due to end-to-end
transmission latency and lower link transmission rates. Depending on your application
profile, this limit might affect cluster performance.
Although each 6780 switch can support up to eight nodes, you should consider the
number of processors supported by each long-distance link between the switches. To
ensure adequate throughput, under conditions of heavy traffic, each link should
support no more than 12 processors. Using all four links between the switches, this
guideline equates to no more than 48 processors for each 6780 switch, regardless of
the number of nodes.
If one of the four links goes down, the other links must handle the traffic for both the
down link and themselves (an additional 12 processors worth of traffic). An OSM alarm
and EMS events are generated when the link is down. If the bad link remains down, it
must be repaired to regain the lost bandwidth.
Note. The long-distance option does not affect the supported lengths for layer-to-layer
connections and switch-to-node connections.