Browse Prior Art Database

Techniques to enhance (double) Oracle OLTP/OLAP performances using unused memory in fail-over nodes in a multi-node CFS cluster (enhanced CFS SmartIO caching)

IP.com Disclosure Number: IPCOM000247123D
Publication Date: 2016-Aug-08
Document File: 4 page(s) / 68K

Publishing Venue

The IP.com Prior Art Database

Related People

Viswesvaran Janakiraman: INVENTOR [+4]

Abstract

CFS based Oracle clusters in A/A, A/P configs have large memory idle in the fail-over nodes. Such clusters also are linked with IB enabling very low latency/high BW remote memory accesses between nodes (RDMA). Our experiments show that we can boost the OLTP numbers from 40 to 100% in a typical configuration by selectively caching the REDO logs with very small buffers and RO data (stok files) with large chunks, in a highly available manner using key SF features.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 44% of the total text.

Page 01 of 4

Techniques to enhance (double) Oracle OLTP/OLAP performances using unused memory in fail-over nodes in a multi-node CFS cluster (enhanced CFS SmartIO caching)

Viswesvaran Janakiraman

Raghupathi Malige

Medha Bhadkamkar

Hao Luo

Abstract

CFS based Oracle clusters in A/A, A/P configs have large memory idle in the fail-over nodes. Such clusters also are linked with IB enabling very low latency/high BW remote memory accesses between nodes (RDMA). Our experiments show that we can boost the OLTP numbers from 40 to 100% in a typical configuration by selectively caching the REDO logs with very small buffers and RO data (stok files) with large chunks, in a highly available manner using key SF features.

Problem Statement

SF-HA and CFS-HA based Oracle clusters in both A/A and A/P configurations waste their precious memory resources in the fail-over nodes as they are provisioned (or over- provisioned) to handle a fail-over condition.

Such clusters typically are also capable of RDMA communication thru IB links. This extra memory could not be used for boosting Oracle OLTP/OLAP performance using Oracle configurations (again party due to fail-over reasons). Symantec has approved patent (S07- 5021) to turn such large pool in A/P configuration into a generic block cache. But such approaches will yield only very limited performance boost based on our experiments. We propose a set of techniques which exploit typical Oracle configuration information with key SF features such as CFS-SmartIO, VxVM mirroring and VCS fail-over mechanisms to boost the performance (experiments show 40% to doubling) a great deal in typical configurations, with high availability.

1

© 2016 Veritas Technologies LLC. All rights reserved. Veritas and the Veritas Logo are trademarks or registered trademarks of Veritas Technologies LLC or its affiliates in the U.S. and other countries. Other names may be trademarks of their respective owners.


Page 02 of 4

Publication Description

Some definitions:

Crash scenarios: (A) active node crash (B) passive node crash (C) both nodes crash

Technique-1: Drastically reduce REDO LOG latency thru very small mirrored write-thru RAM buffer
Oracle OLTP performance is very sensitive to the latency encountered by the REDO thread. These are very small sequential append writes. Even dedicated, high performance disk/SSD have latencies which can be improved by orders of magnitudes by RAM. Our experiments show that a very small (around 100 MB), mirrored, write-thru buffer across the nodes can dramatically enhance the TPM-C numbers. This is done by creating a RAM based buffer device using VxVM over which CFS SmartIO does write-thru caching.

Key aspect is the consistent recovery this model offers exploiting some key features as below:

Crash: (A) VCS will trigger a flush of passive buffer before fail-over starts - no loss (B) VxVM will detach the remote buffer plex - no loss (C) periodic write-thru flush from SmartIO minimizes un-committed writes in small buffe...