Notes from Oracle RAC Customer Summit

Hello Everybody

I had the opportunity this week of assisting to a Real Application Cluster Customer Summit in the new Oracle office of Bridgewater NJ. In the meeting was one of the product managers for Real application Cluster, members of he RACPACK and representation from Oracle Consulting.

I was a very interesting meeting where Oracle was basically asking their non RAC customers what can we do for you to start using it, and to the current customers what can we do for you be more comfortable using it.

They did emphasis on virtualization and the hardware becoming cheaper, the different ways that you can virtualize (server split (VMWARE) or server aggregations (GRID)). They went through the advantages and disadvantages of each one, in summary server split does not offer business continuity, scalability across servers, performance (high overhead) and reduction in the management burden and with the prices of the hardware and they becoming more powerful it almost a no brainier to use grid.

However Server split virtualization has its niche, It is good for systems with smaller workloads, test, development environments and non critical applications where you can justify the price tag involved with using RAC. while GRID works best for larger workloads and business critical applications.

Grid aims to provide dynamic resource provisioning and automation, For Oracle grid is evolving, the first version of grid (GRID 1.0) is the typical 2 nodes cluster database, once their customers start feeling comfortable with the and it's potential they wan to leverage the full potential of grid and evolve into what they call GRID 2.0. Oracle is looking into incorporate to it policy based service level automation.

In the meeting Oracle also spoke about ASM and Clusterware giving us some tips:

- They say that SAN based redundancy is commonly used but when that is not available ASM is a good option.
- They clarified that contrary to what people think ASM stripping in conjunction with SAN stripping does not hurt performance.
- They said there is a benefit in using ASM redundancy when using extended clusters (Clusters separated geographically).
- When using ASM 2 different LUNS should not map back to the same spindle.
- LUN provisioning should be done in a complimentary manner to ASM.
- Redundancy for private network is highly recommended, most customers use a pair of dedicated switches for it.
- Jumbo frames can be used to reduce the private network traffic and CPU overhead (1500 bps to 9k).
- They recommend to use workload management by service and to partition similar workloads by service.