Hi All,

Red Hat Enterprise Linux 5 kernel supports four I/O schedulers:
- cfq (Completely Fair Queuing)
- deadline
- noop
- anticipatory

I read in documentation that the recommended kernel line settings for 64-bit Red Hat Enterprise Linux 5 running as a VMware guest are:

divider=10 notsc iommu=soft elevator=noop

But for single instance databases with dedicated storage the deadline scheduler is recommended. The deadline scheduler reorders I/O to optimized disk heads movement and caps maximum latency per request to prevent resource starvation for I/O intensive processes.

I have an Oracle instance on RHEL5 running as a VMware(ESX) guest with dedicated storage. What scheduler is better in my case?

I would appreciate if you could clarify this...

We have several x86_64 RHEL 5.3 systems that are Oracle servers running under ESX 3.5. All have just "divider=10 notsc". I haven't looked into iommu-- did that come from RHEL or VMware or Oracle? The docs from the different sources seem to say something different every few months, and if nothing is broken I don't obsess over trying to keep up with the latest changes.

I tried an experiment several months ago, probably on RHEL 4.6 and an earlier ESX, and I could not tell any significant difference between the performance of the different I/O scheduler options. When we moved most all of our storage to a fiber-connected SAN, the SAN controller's buffering washed out any kernel optimizations and all the schedulers gave about the same results. I leave it at the default. YMMV depending on specifics of your storage.

Hello,

Thank you for attention to my question. I found about it in Red Hat documentation:
Oracle 10g Server on Red Hat®
Enterprise Linux® 5
Deployment Recommendations
Version 1.2
November 2008

Can you advice me any documentation about Oracle on VMWare?

I have not seen anything special for Oracle on VMware.

I'll take at iommu.-- thanks for the pointer.

You are welcome. Can you tell me about a "noatime" mount parameter for ext2 and ext3 file systems. Is it really improves I\O?

noatime improves I/O by avoiding writes. When it is present the file system manager does not stamp directory and file inodes with the last time they were accessed, which saves the disk writes necessary to update those inodes.

People rarely care about the atime of an Oracle tablespace, so in those file systems this lets you give the system permission to skip that work.