In an Oracle RAC environment, when a block of data is not found in the local cache of an instance, Oracle first checks the cache of the remaining instances in the cluster before reading the block from disk. Cluster latency is the time taken to receive a block from another instance in the cluster.
The charts in the Latency page represent different views of the cluster latency for the Oracle RAC cluster under investigation.
To open the Latency page
Click Cluster| Latency.
Charts on the Latency page
|Cluster Latency||Cluster Latency shows recent cluster latency values (in milliseconds per block) for every database instance within the cluster.|
|Average Ping Time||
A ping request by a user or program causes an Internet Control Message Protocol (ICMP) echo request. If the remote IP address is active, it responds with an ICMP echo reply.
In an Oracle RAC cluster, Oracle uses the cluster interconnect to transfer blocks between instances that participate in the cluster. The average ping time represents the time for a ICMP echo reply to be received back from the network address.
Ping Time can be used to compare the actual echo time versus the cluster latency times displayed by the Cluster Latency chart, and provides a means to diagnose the origin of network performance issues.
Note: The Average Ping Time chart is populated only for (Oracle 10g and later) RAC clusters running on Linux servers. You must also have an OS connection to the host machine.
If the chart still is not populated, verify the ping command package is installed on all nodes in the cluster (installations of the Linux 2.4 kernel may not install the ping command by default). If ping is not installed: (1) install the iputils package on ALL nodes in the cluster, (2) create a new Spotlight on Oracle RAC connection that connects to the operating system (select Monitor OS in the Connection properties dialog: Oracle RAC Connection Details).
|Total Latency||Total Latency shows recent cluster latency values (in milliseconds per block) classified according to latency type, where prepare latency is the time taken to prepare a block, and transfer latency is the time taken to transfer the block across the cluster interconnect.|
When a required block is not found on the local Oracle RAC instance, it must be requested from another instance in the cluster (if available). To do this, the GCS (global cache service) process identified by the LMSn background process needs to:
Prepare Time is the time needed for the GCS to build, flush, pin and send the block to the requester. This chart displays the time taken by the GCS to perform these individual functions on requested blocks. While occasional spikes on this chart are normal behavior, consistent high values indicate that you should investigate one of the following:
|Prepare Latency||Prepare Latency displays the time needed to prepare the blocks requested by another instance in the cluster. The latency is displayed for every instance participating in the Oracle RAC cluster.|
Transfer Latency shows the actual transfer time for the block.
The latency is displayed for every instance participating in the Oracle RAC cluster. The transfer latency also indirectly reflects the performance of the cluster interconnect – a slow transfer time indicates that the interconnect is either overloaded or slow. If not rectified, this will affect the overall performance of the cluster.