UNIX provides a system command, vmstat, to help you monitor system activity such as paging. The abbreviated command syntax is:
vmstat interval count
The interval is the time between rows of output, and count is the number times a row of output is displayed. For more information about vmstat (including its options and field descriptions), see your operating system's documentation. Here is an example:
> vmstat 2 3
procs memory page disk faults cpu r b w swap free re mf pi po fr de sr s0 s1 sd in sy cs us sy id 0 0 0 3312376 31840 0 8 0 0 0 0 0 0 0 0 297 201 472 82 4 14 0 0 0 3312376 31484 2 3 0 0 0 0 0 0 0 0 260 169 597 80 3 17 0 0 0 3312368 31116 0 8 0 0 0 0 0 0 0 0 205 1202 396 67 4 29
The above output shows a steady Sybase IQ querying state where the physical memory of the machine has not been overallocated. Little to no system page faulting is occurring. These next set of examples show vmstat output that indicates a problem. (The output shown omits some of the above fields to fit better on the page.)
procs memory page faults cpu r b w swap free re mf pi po fr de sr in sy cs us sy id 0 0 0 217348 272784 0 148 11 3 9 0 2 251 1835 601 6 3 91 0 0 0 3487124 205572 0 5 0 0 0 0 0 86 131 133 0 1 99 0 0 0 3487124 205572 0 5 0 0 0 0 0 71 162 121 0 0 100 0 0 0 3483912 204500 0 425 36 0 0 0 0 169 642 355 2 2 96 0 0 0 3482740 203372 0 17 6 0 0 0 0 158 370 210 1 3 97 0 0 0 3482676 203300 0 4 10 0 0 0 0 160 1344 225 1 2 97 0 0 0 3343272 199964 1 2123 36 0 0 0 0 213 131 399 7 8 85 0 0 0 3343264 185096 0 194 84 0 0 0 0 283 796 732 1 6 93 0 0 0 3342988 183972 0 17 58 0 0 0 0 276 1051 746 2 4 94 0 0 0 3342860 183632 0 119 314 0 0 0 0 203 1660 529 3 4 94 0 0 0 3342748 182316 2 109 184 0 0 0 0 187 620 488 4 2 95 0 0 0 3342312 181104 2 147 96 0 0 0 0 115 256 260 9 2 89 0 0 0 3340748 179180 0 899 26 0 0 0 0 163 836 531 4 4 92 0 0 0 3328704 167224 0 2993 6 0 0 0 0 82 2195 222 4 7 89
The first line of the above output provides a summary of the system activity since the machine was started. The first three lines show that there is approximately 200MB of free physical memory and that the machine is idle. The fourth line corresponds to Sybase IQ starting up for the first time. Beginning at the eighth line, the amount of free memory starts to reduce rapidly. This corresponds to the Sybase IQ buffer caches being allocated and database pages being read in from disk (note that CPU usage has increased). At this time there is little user CPU time as no queries have begun.
procs memory page faults cpu r b w swap free re mf pi po fr de sr in sy cs us sy id 7 0 0 3247636 58920 0 1880 1664 0 0 0 0 1131 442 1668 80 18 3 18 0 0 3246568 43732 0 709 1696 0 0 0 0 1084 223 1308 90 10 1 12 0 0 3246604 37004 0 358 656 0 0 0 0 600 236 722 95 5 0 15 0 0 3246628 32156 0 356 1606 0 0 0 0 1141 226 1317 91 9 0 19 0 0 3246612 26748 0 273 1248 0 0 0 0 950 394 1180 92 7 0
The above output is from slightly later when the query is underway. This is evident from the user mode CPU level (us field). The buffer cache is not yet full as page-in faults (pi field or KB paged in) are still occurring and the amount of free memory is still going down.
procs memory page faults cpu r b w swap free re mf pi po fr de sr in sy cs us sy id 21 0 0 3246608 22100 0 201 1600 0 0 0 0 1208 1257 1413 88 12 0 18 0 0 3246608 17196 0 370 1520 0 464 0 139 988 209 1155 91 8 0 11 0 0 3251116 16664 0 483 2064 138 2408 0 760 1315 218 1488 88 12 0 30 0 0 3251112 15764 0 475 2480 310 4450 0 1432 1498 199 1717 87 13 0
The above output is from even later. On the third line of the output it shows that the system has reached its threshold for the amount of free memory it can maintain. At this point, page-outs (po field or KB paged out) occur and the level of system mode CPU (sy field) increases accordingly. This situation results because physical memory is overallocated: the Sybase IQ buffer caches are too big for the machine. To resolve this problem, reduce the size of one or both of the buffer caches.
