Tuning Global Cache - Expert Oracle RAC Performance Diagnostics and Tuning - page 390

Service Time
Service Instance Elapsed CPU Time DB Time
Name Name mSec/Call mSec/Call mSec/Call Calls/sec
-------------------- ---------- ------------ ---------- ---------- ---------
DBAUSER PRDDB1 4118982 268000 4118982 .19
PRDDB2 0 0 0 .00
PRDDB3 54175 54500 54175 .39
PRDDB4 0 0 0 .00
PRDDB5 0 0 0 .00
PRDDB6 0 0 0 .00
ENDUSER PRDDB7 7487921 168000 7487921 .00
PRDDB8 756606 60286 756606 3.24
BIAPPS PRDDB1 5683365 287000 5683365 .17
PRDDB2 4879139 358000 4879139 .00
PRDDB3 7232351 287000 7232351 .19
PRDDB4 4280285 161000 4280285 .00
PRDDB5 2176095 174000 2176095 .00
BIETL PRDDB6 1313075 65600 1313075 1.20
PRDDB7 10113853 190000 10113853 .00
PRDDB8 21474104 197176 1474104 3.28
BCKUP PRDDB7 0 0 0 .00
PRDDB8 0 0 0 .00
Step 8
From the output, it can be noticed that instances 7 and 8 are both running services BIETL ; that is probably why these
servers are consuming higher resources. BIETL is a highly insert-intensive data loading application. Because this is an
INSERT mostly application, data sharing between other instances is much lower. This was noticed from the extremely
low activity across the interconnect on other nodes in the cluster. If there are a sufficient number of servers and data
can be further distributed, the additional resources required could be obtained by adding additional nodes during the
time of peak load and balance resource utilization.
Think Outside the Interconnect
When a user process retrieves data from the database connecting to one instance in the cluster, the data is usually
cached locally on the instance. Subsequently, when another user process on another instance requires the same
block of data and is not able to find a buffered copy in its local cache, it can acquire the block image from the holding
instance over the interconnect faster than reading the block from disk. Cache Fusion moves current blocks between
instances (illustrated in Figure 13-1 ) rather than re-reading the blocks from disk. When a consistent block is needed
or a changed block is required on another instance, Cache Fusion transfers the block image directly between the
affected instances.
Why is that so? Why is reading blocks from disk unfavorable to reading over the interconnect? This is because the
latency of performing a single physical disk I/O is much higher compared to the latency of reading or transferring a
block of data over the network. On the contrary, if the latency of the interconnect is higher compared to the latency
of the disk system, it defeats the purpose of implementing a RAC solution. Argument against this could be that with
recent developments with improved high-speed storage, this may not be true anymore. However, although there
have been considerable improvements in the performance of the storage subsystems, equally there have been