Database Reference
In-Depth Information
We have found the components involved in prepare latency. What about interconnect latency? The average
interconnect (roundtrip) latency depends on the number of data blocks received and the time required to receive
these data blocks. Average interconnect latency can be computed using the following:
Average interconnect latency = Blocks receive time ÷ blocks received
Data blocks received is the sum of both the current blocks received and the CR blocks received. Similarly, the
time taken to receive these blocks is also the time taken to receive a current block and the CR block.
Once again, all the instrumentation required to compute the transfer latency is also available with the Oracle
database in the
GV$SYSSTAT
and
GV$SESSTAT
views. Data required to compute the transfer latency could be arrived at
using the following two equations:
Blocks receive time = (gc cr block receive time + gc current block receive time) × 10
Blocks received = (gc cr blocks received + gc current blocks received)
The statistics would point to either the prepare phase or the transfer phase. If the statistics on the transfer
phase are significantly on the higher side, causing concern, for obvious reasons, the interconnect configuration and
setup may have to be looked at. However, if the prepare phase is of concern, the root cause of the problem should
be identified by looking at what is causing these high numbers. The reasons could be resource contention, lack of
resources, or the process is simply waiting or hung.
■
Detailed discussions on the transfer phase with respect to the interconnect and the components involved in
the transfer phase are forthcoming in Chapter 14.
Note
Workshop
We now try to understand the environment illustrated in Figure
13-6
. The eight-node cluster is configured into
two server pools: Server Pool 1 (
SSKYPOOL1
), which is configured with BIAPPS uniform service, and Server Pool 2
(
SSKYPOOL2
) are configured with
BIETL
uniform service.
Search WWH ::
Custom Search