vmstat 1 16 :
Legend:
- kthr = kernel thread - state change (the columns selected are of interest here)
- r = run-queue - number of tasks executing and consuming CPU resources - average per interval
- b = blocked-queue - number of tasks waiting (blocked ) for CPU resources - average per interval
- pi = pages-in - memory contents read to memory from disk (where it was paged out) - avg per second
- po = pages-out - memory contents written from memory to disk (non-zero is NORMAL) - avg per second
- wa = waiting-CPU - percentage of CPU time waiting for external operations (I/O) to finish - avg per interval
A few simple rules of thumb tell us that, when busy, the system is:
MEMORY-BOUND: - the paging size (in MB) > SGA, or non-zero values in the pi column, or page percent too high
CPU-BOUND: - the average values in columns r and b > # of CPU's, or load max > load coef (e.g., 10)
I/O-BOUND: - there are values in the wa column > 20% (up to max 40 percent by some authors)
RUNS-AT-FULL-CAPACITY: - the sum of us and sy approach 100% and id and wa are very small
