Re: 12c pluggable database shared SGA question

From: Job Miller <jobmiller_at_yahoo.com>
Date: Tue, 16 Sep 2014 13:08:09 -0700
Message-ID: <1410898089.65546.YahooMailNeo_at_web140403.mail.bf1.yahoo.com>



For those that don't read the entirety of the white paper the summary section identifies the following efficiency gains from Multi-tenant based on some testing that was done.

http://www.oracle.com/technetwork/database/multitenant/learn-more/oraclemultitenantt5-8-final-2185108.pdf

Overall Summary

The tests presented in this white paper demonstrate the efficiency gains from sharing of background processes and memory in Oracle Multitenant. The aggregation of work in background processes across pluggable databases saves a significant amount of CPU time and allows to reduce the number of cores needed to support 252 active OLTP databases from 196 to 128 cores.

Aggregation of redo log and data block flushing across tenants further reduces the requirement on storage I/O from 356,000 IOPS to 131,000 IOPS.

The sharing of SGA pools and background processes reduces the memory footprint of a database significantly and saves a total of 368 GB memory, which either translates into hardware cost savings, or can be reinvested by increasing the buffer cache of the CDB.

Dynamic workloads benefit further from the on-demand allocation of buffer cache, which helps them to accommodate load peaks that they might not be able to satisfy with a static resource allocation.

Write-intensive workloads will benefit from the aggregation of work in database writer and log writer processes, which results in lower CPU consumption for background processes and fewer disk writes.

The saved CPU resources are available for more SQL processing instead. While individual log writer processes in non-CDBs might achieve lower log file sync times, they only accomplish this at the cost of a high rate of disk writes, which requires a very high-performing storage. With a slower storage, those wait times could be similar or even higher for non-CDBs than for PDBs.

Read-intensive workloads benefit from the reduced memory footprint of PDBs, which allows to consolidate more of them on the same amount of hardware. Since they commit less frequently, their response time is mostly affected by SQL processing and buffer cache efficiency rather than log file sync, and can be even better with PDBs than with non-CDBs due to an overall large buffer cache.

Workloads with more complex SQL, even if highly transactional, will be more impacted by SQL execution time instead of commit time. Since SQL execution can be more efficient in Oracle Multitenant with less disturbance from background processes, any potential increase in log file sync times might be more than compensated for by faster SQL processing.

Workloads with dynamic behavior can largely benefit from on-demand allocation of buffer cache, CPU time, and other resources in Oracle Multitenant, whereas single instances or virtualized deployments allocate many of these resources statically, which can lead to large over-provisioning and low resource utilization


--
http://www.freelists.org/webpage/oracle-l
Received on Tue Sep 16 2014 - 22:08:09 CEST

Original text of this message