Which OS parameters should be modified after addon RAM?

[ericfjchen]

This is ES80 with Tru5.1B. After addon RAM from 10G to 48G, which OS parameters(/etc/sysconfigtab) should be modified besides enlarging swap space?

[Ivan Ferreira]

It depends of your system, but normally, these parameters should be modified:

max_per_proc_address_space
per_proc_address_space
max_per_proc_data_size
per_proc_data_size

These values normally are equal to the amount of memory of the system.

[Mark Poeschl_2]

You may also want to consider enlarging or adding to your swap space. See output of 'swapon -s' for your current configuration.

If you are running a database or other application that makes heavy use of shared memory regions you might also want to adjust some of the 'ipc' kernel parameters - in particular shm_max. This should be accompanied by adjusting database/application settings that control shared memory usage.

[Martin Moore]

Another possible consideration is whether you're running granularity hints (GH) for shared memory, e.g., if you're running a database system. On a NUMA system such as the ES80, this should be configured with the rad_gh_regions[] parameters in sysconfigtab. If you want to increase the size of your pre-allocated GH memory, say if you're increasing the size of an Oracle SGA, don't forget to adjust these parameters as well.

The swap space point above was a good one. If you're running in eager swap mode (the default), you may well need to add more swap. (If you're running in lazy mode, it won't matter.)

Martin

[Han Pilmeyer · Accepted solution]

You first need to tell us what you want to use the additional memory for. Note that the "per_proc" parameters that Ivan mention are defining limits for virtual address space of processes. Typically you do NOT want to set those to the size of your physical memory, unless you're using all of the systems memory for process that need to address all of the system memory. Databases typically don't do this.

If you're running a databases and you're adding more instances or more processes per instance, then the "per_proc" parameters can probably stay the same. However if you're increasing the SGA size, then things like the number and size of shared memory segments, the optional granularity hints memory, etc. may need to be changed.

maxusers is one parameter that one could change. Although it isn't really necessary as it can be dynamically increased.

In short, tell us what you want to use the memory for and we can give better advice.

[ericfjchen]

1. How do I check whether we are running granularity hints (GH) for shared memory?

2. There is only one Oracle database(one instance) on this box. We would like to improve the database performance by enlarging the memory. So which OS parameters should we change?

Thanks

Eric

[Manish PATHAK_2]

I am agree with Ivan,normally you need to chang only these values.

max_per_proc_address_space
per_proc_address_space
max_per_proc_data_size
per_proc_data_size

Also depends on what processes you are running.

All the respetive values you can find in the Administration manual (normally equivalent to RAM size).

Regards
Manish PATHAK

[Han Pilmeyer]

If you're using granularity hints on an EV7 based system, you will have set the rad_gh_regions parameters to a non-zero value.

If you want to use the memory to make Oracle run faster (not more clients connecting), then you can increase the SGA size and possibly the PGA (sort areas).

If you decide to increase the SGA, you need larger shared memory regions. ipc:shm_max defines the largest shared memory size. ipc:shm_mni the maximum number of those shared segments. These two multiplied will have to be able to map the entire SGA and any other shared segments you use. When you have a very large SGA, using granularity hints will be more efficient (you have to change ipc:ssm_threshold and vm:rad_gh_regions[X] to enable them).

{,max_}per_proc_address_space will need to be large enough to hold both the SGA and PGA. Note that if you decide to increase the PGA that this will increase the memory usage of every Oracle process.

You have a lot of additional memory. What's the size of your SGA now and how many Oracle processes do you have?