Tuning the Cluster Interconnect - Expert Oracle RAC Performance Diagnostics and Tuning - page 483

Figure 14-14. Cluster cache coherency

Latencies on the cluster interconnect could be because:

• No dedicated interconnect for cache fusion activity has been configured

• A large number of processes in the run queues waiting for CPU or due to processor

scheduling delays

• Incorrect platform-specific O/S parameter settings that affect IPC buffering or process

scheduling

• Slow, busy, or faulty interconnects

The packet rates being transferred over the interconnect can also be determined using the ping command and
can be generated by the AWR global report 8 (Figure 14-15 ) or using the tcpdump command from the O/S.

tcpdump -ni eth24
17:49:08.019260 IP 169.254.100.30.65166 > 169.254.73.36.54220: UDP, length 192
17:49:08.019561 IP 169.254.100.30.55896 > 169.254.73.36.15879: UDP, length 8328
17:49:08.019585 IP 169.254.100.30.41485 > 169.254.73.36.50818: UDP, length 192
17:49:08.019721 IP 169.254.73.36.10368 > 169.254.100.30.49351: UDP, length 192
17:49:08.020410 IP 169.254.100.30.55896 > 169.254.73.36.16941: UDP, length 192
17:49:08.020679 IP 169.254.100.30.55896 > 169.254.73.36.18966: UDP, length 8328
17:49:08.021046 IP 169.254.100.30.49351 > 169.254.73.36.29081: UDP, length 440
17:49:08.021126 IP 169.254.73.36.13329 > 169.254.100.30.43241: UDP, length 5768

8 This is generated using $ORACLE_HOME/rdbms/admin/awrgrpt.sql .