hello everyone. after exhausting all reading material on multiple instances of sql 2k on a cluster I wanted to run my config by anyone that would listen to determine if I got it right. I have a 2 node active/active cluster with 6GB ram on each node. this is a dedicated sql cluster and each instance on each node will run one db. from what i've read I need to update boot.ini on each node with the /3GB and /PAE switch and then make sure that the sum of the max server memory for each node (in case of failover) does not exceed the physical ram in either of the 2 nodes. That being said I would think it would be safe to give each instance on each node 2.5GB as a max server memory. If one node fails that would mean the other would run 2 instances @ 5GB total leaving 1GB for server processes and managing AWE. Do you think this config is a good starting point. I have not tested it yet but will start next week. thanks in advance for any comments or suggestions. Again this cluster is dedicated SQL only. not other apps.
My understanding is that in a Windows environment without AWE and /3GB that the Operating system will take up about 2GB. Thus in a 4GB system, the OS get 2 and SQL would get 2. By setting the /3GB you can constrain the operating system to using 1 GB and SQL would get 3GB. It is also my understanding that the management of AWE takes a significant amount of memory.
Given the scenario you describe limits the operating system and the management of AWE to 1 GB. I suspect that this amount of memory will not be sufficient. I can not now remember the details, but in one situation I read that the management of AWE required 1 GB (I think this was on an 8 GB system).
I think Solart may be right. It is hard to say in which situations what is the best memory config. My suggestion to you is that you test each possible config, and go with the best. There are situations when PAE and AWE cause performance degradation.
yeah I guess you are both right. What sucks is that the examples that books online gives you and other documents on the web are for systems that have 4GB, 8GB, 16GB, 32GB etc etc. There is no matrix which tells you best starting points with memory sizes...like mine 6GB. The documents are pretty vague at times, I guess the only way us guy's know is with plenty of testing.....let me know if you think of anything else. I'll keep you posted on my testing.
quote: I guess the only way us guy's know is with plenty of testing
And that's the way it should be. Documentation such as this should be used as suggestions, starting points etc... You should always test because no document can know your exact situation. Testing is always the recommendation.