Database Reference
In-Depth Information
Solaris
bring the interface down (unplumb) and set the instance to accept jumbo frames:
# ifconfig <interface> down unplumb
# ndd -set /dev/<interface> instance 1
# ndd -set /dev/<interface> accept-jumbo 1
# ifconfig <interface> plumb <address> up
■
Best Practice
Jumbo frames provide overall performance improvements in a raC environment and should be used.
Jumbo frames only help lan performance; traffic leaving the lan to the internet is limited to packets of 1,500 bytes.
access to and from a raC environment is mostly limited to the application server and local clients setting up jumbo
frames should provide positive benefits. large MtU sizes between the application server and database servers will allow
increase of the session data unit (sDU) sizes. this allows for transfer of large volumes of data between the application
server and database, such as when using array fetches.
More discussions on sDU and buffer sizing are forthcoming in Chapter 11.
ASM
The storage system is a shared component in the RAC architecture and although compared to a single instance
configuration, its usage is of equal importance and performance is critical as a shared component of the cluster.
Initial scalability testing of the storage system should be performed during capacity planning to determine the
write storage allocation and best size and number of disk combinations. Scalability testing should be done taking
into consideration the LUNs or individual disks (in case “just a bunch of disks” [JBOD] is used for storage), and I/O
performance characteristics should be determined to ascertain performance. Tools such as Oracle Orion can be used
to benchmark the I/O system without having to use an Oracle installation.
Instance
The scalability of an instance in the cluster depends largely on the quality of the executing SQL and how the memory
pools are sized. The memory used by an Oracle instance is divided into shared pool, large pool, buffer cache, redo log
buffer,and so forth. The sizing of these memory areas is influenced by parameters such as the degree of parallelism,
multiblock read count, and so forth.
Almost always scalability limitations in clustered environments are caused by SQL statements that are not tested
for memory efficiency and performance as well as memory caches that are not sized right.
With the dynamic provisioning of instances using the policy managed database option, scalability can be
obtained by provision instances to the pool based on demand if the utilization of CPU and memory of a system cannot
provide good service.
SQL*Net
System scalability also depends on the network and NIC between the application server and database tier. What is the
packet rate that an interface supports? What packet sizes can be supported for best performance? The configuration
Search WWH ::
Custom Search