180 likes | 514 Views
Wim Coekaerts Director of Linux Engineering Oracle Corporation. Oracle Cluster Filesystem on Linux One year Later. What is a CFS ?. Shared Disk Cluster Filesystem Improve management of data for Oracle Real Application Clusters (RAC) Customer demand for a CFS to run RAC
E N D
Wim CoekaertsDirector of Linux Engineering Oracle Corporation
What is a CFS ? • Shared Disk Cluster Filesystem • Improve management of data for Oracle Real Application Clusters (RAC) • Customer demand for a CFS to run RAC • Open solution on the Operating System side (Linux) free and open source
OCFS Version 1 Releases • Beta Release August 2002 • Open source project (GPL license) • Production October 2002 • Red Hat Advanced Server 2.1 for IA32 • Production February 2003 • United Linux 1.0 for IA32 • Production August 2003 • Red Hat Advanced Server 2.1 for IPF/IA64
What did we solve ? • 255 raw device limit on Linux • Difficult to manage raw disk • Single view of data files across all nodes • Use of standard OS tools (modified) to handle backups
What didn’t we solve yet ? • Shared Oracle Home installation • Binaries and generic file support on CFS • Support for other applications (ISV applications) • Volume management • Dynamic resizing of partitions • Defragmentation of filesystem • Online FSCK
Customer Experiences • Large number of production customers today • Easy installation and configuration of CFS • IO throughput is equivalent to RAW IO • Use of the default OS tools can cause slow throughput and kernel issues. • After a few months more requests for generic filesystem support
Do’s and Don’ts • Red Hat Advanced Server 2.1 for IA32 • Use 2.4.9-e.24 or higher kernel • Buffer cache management • Red Hat Advanced Server 2.1 for IPF • Use 2.4.18-e.37 or better • United Linux 1.0 for IA32 • Use Service Pack 2a or better
Do’s and Don’ts (2) • Updated fileutilities from http://oss.oracle.com/projects/ocfs • Dd, cp, tar, md5sum, cksum, mv • O_direct=yes • Always use the above utilities with the O_DIRECT option for better performance and reliability • Use RMAN if possible
Do’s and Don’ts (3) • Use a large blocksize for IO • Avoid small allocations (use 1MB and larger) • Separate archive log files from datafiles • If possible separate out in separate directories • Spread load of files across number of volumes (don’t use 1 volume for everything) • Having one volume causes IO bottlenecks • Use large autoextend size
Do’s and Don’ts (4) • Use comm_voting=1 in /etc/ocfs.conf • Async IO support in OCFS 1.0.9 with specific kernels (need patch for oracle 9.2.0.4) • Mix IA32 and IA64 nodes • OCFS is compatible between the nodes • Easy move of Oracle between architectures (30 seconds)
Customer story • Amazon implemented OCFS for a system • Grant McAlister’s experiences
What is coming next ? • OCFS Version 2 • Shared Oracle Home • Binaries and logfiles • Faster non database operations • Other platforms • Port to AMD64 for version 1 in progress • Port to PPC64 for version 1 in progress
Q & Q U E S T I O N S A N S W E R S A
Next Steps…. • Recommended sessions • Performance and Scalability of Linux in the Enterprise • Tuesday 11:00 am Jamshed Patel • Improving information Assurance of Linux • Tuesday 5:00 pm Mary Ann Davidson • Visit the DEMOgrounds for a customized architectural review, see a customized demo with Solutions Factory, or receive a personalized proposal. Visit the DEMOgrounds for more information. • Relevant web sites to visit for more information • http://oss.oracle.com/projects/ocfs
Reminder – please complete the OracleWorld online session surveyThank you.