This content has been marked as final. Show 5 replies
user8803475 wrote:Recommended? 2.
1. How many voting disks do we need per cluster (considering we have a redundant Fibre-channel connections)?
Voting disks should be mirrored.
2. How many OCR disks do we need and what is the recommended access (OCFS2 or RAW)?Recommended? 3.
OCR disks should be multiplexed. And on RAW disks as ocfs2 introduces another s/w layer and increases CRS dependency and complexity.
3. How do we calculate the bandwidth needed for the cluster interconnect? We currently have configured 2 x 1Gbit Ethernet ports for this.Not possible. You cannot accurately thumbsuck what loads will be put on the Interconnect. The Interconnect has two primary requirements - performance and scalability.
And 1Gb ethernet ports are neither.
Any thoughts?Well, seems like you are asking the very same questions under different guises here and on the RAC forum. And that I do not appreciate at all....
Ask once. If you want more details, bump the thread, or phrase your questions differently. But asking these under different guises is not going to change the default answers.
I discovered that this DBA is somebody else that have the same interview questions, the questions is emailed , that is why we are typical :)
sorry again , but this is really funny
expect the same 3 exctly questions to be asked again from another DBA
see more Oracle RAC interview questions and answers at ..
Shouldn't it be the other way.
1. How many voting disks do we need per cluster (considering we have a redundant Fibre-channel connections)? Recommended? 3. Voting disks should be multiplexed. 2. How many OCR disks do we need and what is the recommended access (OCFS2 or RAW)? Recommended? 2. OCR should be mirrored.
The critical thing with the cache interconnect is not bandwidth ... it is latency. Again GigE is not the best possible solution. But if the application is properly written and designed around node affinity and services it shouldn't be that big a deal though in the real-world it is.
What I find fascinating is that the single most important aspect of RAC I see done incorrectly in database after database including at some of Oracle's largest customers is the creation of services using DBMS_SERVICE. If someone does not understand the difference between CLB_GOAL_SHORT and CLB_GOAL_LONG I don't see how they can be successful.