|
|
| Resource model distribution |
|---|
|
This resource model should be distributed to Windows endpoints. |
The Memory resource model highlights the following problems:
When cache is used, four different areas are examined:
The Hit Percentage Rate is examined for each area. This gives an idea of how much data was supplied by the cache and how much was supplied from another location. If cache is used, ideally it should provide at least 70% of the data. A lower percentage indicates data is being lost in the cache or that the cache is not large enough.
Committed bytes indicate the amount of virtual memory, in bytes, that have been committed. The commit limit is the current limit of physical space (either RAM or in the pagefile) that is available for the committed bytes.
If the committed bytes become larger than the commit limit, the pagefile must be enlarged. Having a pagefile that is constantly in flux causes some performance problems because the system spends time extending the pagefile or shrinking it. This process also involves the slowest part of the system and is prone to becoming a bottleneck.
The committed bytes has an upper limit. This is dictated by a setting in the system control applet. If the pagefile reaches its maximum limit, it is almost guaranteed that available memory will be extremely low. In this situation a major performance degradation can be expected.
If committed bytes continues to grow and there is not enough available physical space to hold the committed memory, a core dump will probably occur.
The initial size of the pagefile is best determined by the amount of RAM in the system. A good guideline is 150% of the amount of RAM.
Available memory is checked constantly to ensure it does not drop below 10 MB. When available memory is low, performance declines slightly, but the system still works correctly. However, the administrator needs to know that memory is low and may need to monitor it.
As available memory drops, Windows NT systems try to keep the available memory above 4 MB. At this point memory is too low and the operating system spends more time keeping memory available than processing requests. Additionally, high paging begins to occur, and the pagefile reaches its maximum size.
These three different areas of memory need to be analyzed for a possible leak:
The top five processes are examined in the Private Bytes. The total pools of System Code and System Drivers are also examined for growth. This generally indicates a memory leak in the affected area of memory.
Page Faults occur when Windows NT moves a piece of data to another portion of RAM or to the pagefile. Paging moves data to the hard drive, so it is incorporated in the counters for Page Faults.
Page faults are considered high when they reach approximately 350 page faults per second. Paging is considered high at approximately 60 pages per second. Performance declines because the system moves data to and from the hard drive, typically the slowest component of the system. The hard drive and processor become busier and overall performance of the machine decreases.