Tuning the Cluster Interconnect - Expert Oracle RAC Performance Diagnostics and Tuning - page 463

INterCONNeCt traFFIC tICKetS

as more and more instances are allocated to the cluster, the number of tickets allocated to every instance
diminishes, thus hindering the actual number of tickets available to the instance. This indirectly causes contention
on the packets being sent across the interconnect. The amount of buffer allocated is shared by

• asM

• Database

Table 14-2 illustrates the ticket allocation from the pool of tickets defined by the _lm_tickets parameter
(default value of 1,000). The table illustrates how the number of tickets available is reducing every time a new
instance is added to the cluster. in a single-node cluster, there is no interconnect traffic, so no tickets are used or
allocated. When the second node/instance is added to the cluster, the 1,000 tickets are used by both instances.
subsequently, starting with the third instance, the number of tickets reduces to 500 per instance and reduces
further as more and more instances are added to the cluster.

Table 14-2. Diminishing Allocation of Tickets

Number of Instances Tickets Allocated Tickets Available

1 0 0

2 1,000 750

3 500 375

4 333 250

5 250 188

6 200 160

7 166 125

8 142 107

Tickets allocated = _lm_tickets/ (#number of instances -1)

Tickets Available = 0.75 * tickets allocated

applying the preceding formula to the scalability of the cluster, the _lm_tickets should be increased (so should
the buffer size) if the cluster has more than 8 instances in the cluster. The graph in Figure 14-6 illustrates how the
ticket allocation reduces with the addition of instances to the cluster.