Job Statistics: vmstat

The command vmstat produces information about virtual memory, CPU usage, and disk usage. This is useful for determining if your system is properly configured and if your programs are paging extensively. The command vmstat takes two arguments, one for how long to wait between reports and the second for how many reports to write out before quitting:   

% vmstat 1 8   	Write a report every second for 5 seconds.   
procs  memory           page             faults        cpu   	  
-- --- -------- ---- ----   	   
r  b avm   fre re  pi  po  fr   sr  cy  in  sy  cs us sy id wa   	   
1  0  4680 365  0   0   0   1    3   0 123 102  75  0  2 97  1   	   
1  0  4680 364  0   0   0   0    0   0 121  86  67  0  3 76 21   	   
1  0  4680 364  0   0   0   0    0   0 114  70  28  1  0 99  0   	   
1  0  4680 364  0   0   0   0    0   0 123  70  32  2  0 98  0   	   
1  0  4680 364  0   0   0   0    0   0 114  74  28  1  0 99  0   	   
1  0  4680 364  0   0   0   0    0   0 113  84  67  0 21 69 10   	   
1  0  4680 364  0   0   0   0    0   0 127  78  56  8  9 82  1   	   
1  0  4680 364  0   0   0   0    0   0 119  91  71  0  3 86 11   	   

The actual form of the output may vary somewhat from this, and again you should issue the man vmstat command to get the details for your system.

To help understand this output, we examine some of the fields.

• The first two columns under procs show the number of processes on the system.
  • Column r shows ``runable'' or active processe.,
  • Column b shows ``blocked'' processes awaiting a resource such as input/output.
• Under memory:
  • The avm column under memory denotes active virtual memory,
    • that is, memory which resides on the slow, hard disk and not in the fast electronics. The number in the avm column is the number of pages, usually of 4 MB size, being used by that process. This is a good indication of how much memory your system is using; if this number is significantly larger that the size of RAM in your computer, then the system may be spending too much time reading data onto and back from the hard disk (paging). In these cases performance could be increased by adding more RAM.
  • The fre column under memory in the listing is the number of pages, again of 4 MB size, of RAM that is unused.
    • This is, however, an unreliable measure of free memory since the table of free pages is not updated until the system needs more pages.     

The columns in the vmstat listing under cpu indicate how the CPU is spending its time. The numbers are the percentages of CPU time for the past interval spent on:

• us ``user time,''
  • that is, doing things like running calculations, manipulating data, and drawing graphics and the normal activities of user programs;
• sy ``system time,''
  • that is, kernel activities such as input/output;
• id ``idle time,''
  • that is, something we all wish we had more of;
• wa ``wait time,''
  • that is, the time spent waiting on a device to be ready, usually for input/output. A system well matched with its CPU should spend most of its time in users' work since they are the ones paying the bills. While programs containing lots of I/O can push the system time up, in normal use it should be low, say below 33%, even when the system is transferring large files.

It is illuminating to start vmstat before submitting a large, numerically intensive program and then redirect vmstat's output into a file you can study later. As you observe the reports issued by vmstat every few seconds, you should be able to see how well the system preformed while running your program and possibly spot problems like excessive paging or input/output.

---

Next: I/O Statistics: iostat Up: Managing the System Previous: Processes: ps

---

---