The CPUs / Memory Detail screen provides the detail of the physical memory of each processor.
Physical memory refers to what was installed in the CPU, such as a 64 Gigabyte processor (note that some platforms do not provide all physical memory). Swap (swappable) memory is available for virtual memory and could be swapped to disk (a user process would normally occupy virtual memory space). Free memory is not allocated. Locked memory is forced to remain in the processor and may not be swapped to disk (i.e. virtual memory). Lockable memory is the maximum amount that can be locked. Some system processes, control blocks, I/O buffers, and dirty cache are examples of memory usually locked.
The values displayed in Memory in Megabytes section are originally reported by the Operating System as a number of memory pages which are then converted to Megabytes as follows:
(nbr-pages * bytes-per-page) / bytes-per-megabyte
Where -
bytes-per-page = 16384
bytes-per-megabyte = 1048576
and may be converted to Gigabytes taking the displayed value and divide it by 1024. The value bytes-per-page is obtained from the operating system.
Page Faults occur when a process needs to access memory but it is not present. The System must either read the memory from disk or create a new page.
CME Events (Correctable Memory Errors) is the situation where a hardware check-sum algorithm indicates a memory bit was in error but the System could recover from the event. CMEs are usually transient in nature. The value reported here comes from the Guardian procedure call PROCESSOR_GETINFOLIST_.
HCME Events (Hard Correctable Memory Errors) is a situation on newer NonStop platforms where a CME memory errors exist but has exceeded an internal threshold triggering for removal the memory page where the error is located. The problem does not necessarily indicate a hardware failure. The value reported here comes from the MEASURE CPU Entity.
UCME Events (Uncorrectable Memory Errors) is the situation on older NonStop platforms where a hardware check-sum algorithmindicates a memory bit(s) was in error but the System could NOT recover from the event. If the UCME located in an unused area of memory is 'locked out' from further usage. If the error is located in a User process the operating system may cause the process to abend. If the error is located in a system process or system data area it may cause the operating system to halt the CPU. The value reported here comes from the MEASURE CPU Entity.
Memory pressure is an indication of the need for memory. This value ranges from 0 to 7, with 0 being no pressure and 7 being extreme pressure. Observation of this value indicates it is an average over time that increases rather rapidly but decreases rather slowly.
Memory queue length is the number of processes waiting for memory. Newer platforms do not see a value in this field because memory service occurs within the context of the process itself.