On PostgreSQL 8.1, while a long vacuum is running, the output of
vmstat 10 looks like this (sorry, can't format this very will in this
e-mail client):
r b swpd free buff cache si so bi bo
in cs us sy id wa st
5 2 112 53732 4388 1163404 0 0 13524 1322 941 75053 45
55 0 0 0
3 2 112 51596 2336 1160032 0 0 20731 732 999 196116 35
58 0 7 0
3 2 112 53784 2448 1155388 0 0 5729 1680 716 114442 46
52 0 2 0
7 1 112 52088 2568 1156796 0 0 205 1919 546 28929 56
44 0 0 0
3 2 112 51772 2652 1157028 0 0 162 2010 534 2845 56 44 0 0 0
3 2 112 52296 2736 1156332 0 0 174 1910 517 2648 53 47 0 0 0
2 2 112 52736 2832 1155388 0 0 173 1960 521 2698 57 43 0 0 0
6 1 112 51172 2908 1157400 0 0 173 1825 511 2621 52 48 0 0 0
2 1 112 52156 2984 1155876 0 0 152 1766 495 2669 53 47 0 0 0
5 2 112 53828 3064 1154028 0 0 124 1762 498 2673 54 46 0 0 0
5 0 112 53396 3140 1154684 0 0 122 1803 532 2871 52 48 0 0 0
6 1 112 52004 3240 1155892 0 0 150 2004 522 3712 56 44 0 0 0
5 2 112 53100 3348 1152364 0 0 128 2142 543 2711 54 38 0 8 0
1 2 112 56704 3428 1148132 0 0 102 2229 560 2925 56 44 0 0 0
2 1 112 54876 3500 1150080 0 0 91 2129 598 3068 65 35 0 0 0
3 1 112 53196 3588 1152164 0 0 90 2091 542 2803 57 43 0 0 0
8 2 112 53528 3664 1151868 0 0 106 2180 553 2947 55 39 0 6 0
Vacuum hasn't stopped yet at this point - its been running for many minutes.
On PostgreSQL 8.3, we had this:
r b swpd free buff cache si so bi bo
in cs us sy id wa st
6 0 9604 51932 14276 1138996 0 0 3 1808 523 5115 45
9 42 4 0
2 0 9580 53284 14124 1138092 10 0 564 2561 585 5126 53
10 19 18 0
7 0 9564 53412 14144 1137696 0 0 0 1682 497 4985 42
9 48 2 0
3 0 9532 53320 13880 1137940 6 0 37 1757 538 4979 48 8 41 3
More random data:
On an overnight test of PostgreSQL 8.3 on Fedora Core 6, with
auto-vacuum on (using all default settings) the amount of time that it
takes to run a manual vacuum on a constant sized database (only row
updates, or deletes followed by a replacement add) has doubled from 2
seconds to 4 seconds. A reindex brings the time back down to 2
seconds.
Dan
