Greetings: Kswapd on my SLES 8 whipsaws one of my CPU's to 99.9% for prolonged periods of time during high I/O. While this activity is happening, performance and interactive response time goes by way of the swamp. Any ideas on a workaround? Or shall I call for an exorcist? :) Kernel: 2.4.21-185-smp for AMD64 # vmstat 15 procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu---- r b swpd free buff cache si so bi bo in cs us sy id wa 1 3 312120 5867896 35876 19916 3 2 287 145 4 3 9 0 90 0 1 3 312168 5877016 36220 11832 764 716 16148 728 1294 1761 1 28 72 0 1 6 312228 5867660 36036 18520 2512 316 19864 316 1318 1853 1 28 71 0 1 3 312196 5876848 35836 12200 1084 1972 20998 2026 1887 1901 1 29 70 0 1 3 312180 5875688 36044 13548 728 440 16824 454 1165 1672 1 26 73 0 1 4 312188 5873016 35956 15848 1176 308 21140 308 1462 1852 1 27 71 0 2 2 312168 5878064 35868 11228 1438 598 20180 636 1471 1789 1 27 72 0 1 4 312196 5870220 35708 18872 1502 278 23968 278 1382 1892 1 28 71 0 1 3 312196 5870716 35888 17908 1494 472 25568 484 1683 2038 1 28 71 0 2 4 312188 5875772 35784 14068 998 570 18276 606 1217 1783 1 27 72 0 1 3 312200 5870384 35724 18780 1698 316 30080 316 1957 2185 2 29 69 0 1 4 312192 5871160 35940 17648 1572 676 24746 688 1771 1931 1 29 71 0 2 3 312184 5875620 35964 13664 848 466 16980 520 1249 1702 1 27 72 0 1 7 312216 5869276 35912 19860 1544 316 22562 316 1379 1845 1 28 70 0 1 4 312160 5873284 35516 13440 2870 618 27232 632 1987 2215 5 29 66 0 1 3 312200 5874528 36096 14328 1252 1752 18698 1798 1471 1654 1 27 72 0 2 3 312220 5872704 35808 15972 1242 646 27400 658 1942 1959 1 29 70 0 1 4 312196 5875308 36096 13628 874 506 16932 518 1208 1655 1 28 71 0 TOP Tasks: 100 total, 2 running, 98 sleeping, 0 stopped, 0 zombie Cpu(s): 0.7% user, 27.8% system, 0.0% nice, 71.5% idle Mem: 7873748k total, 1995220k used, 5878528k free, 36072k buffers Swap: 4200956k total, 310804k used, 8091108k free, 10976k cached PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 11 root 25 0 0 0 0 R 99.9 0.0 106:14.98 kswapd 9409 oracle 15 0 1586m 1.4g 1.4g D 4.2 19.2 42:20.19 oracle 6204 oracle 15 0 1160m 1.1g 1.1g D 3.9 15.0 0:37.22 oracle 5488 oracle 15 0 967m 962m 962m S 2.3 12.5 0:31.33 oracle 9407 oracle 15 0 1568m 1.4g 1.4g D 1.6 18.9 37:41.11 oracle 6256 oracle 16 0 456 304 276 R 0.6 0.0 0:00.26 top 9426 oracle 21 5 4880 0 0 S 0.3 0.0 0:00.05 dbsnmp 9439 oracle 20 5 4880 0 0 S 0.3 0.0 0:03.61 dbsnmp 9440 oracle 20 5 4880 0 0 S 0.3 0.0 0:09.36 dbsnmp 6173 oracle 15 0 192m 190m 190m S 0.3 2.5 0:06.96 oracle 1 root 15 0 48 0 0 S 0.0 0.0 0:14.94 init 2 root RT 0 0 0 0 S 0.0 0.0 0:00.00 migration_CPU0 3 root RT 0 0 0 0 S 0.0 0.0 0:00.00 migration_CPU1 4 root RT 0 0 0 0 S 0.0 0.0 0:00.00 migration_CPU2 5 root RT 0 0 0 0 S 0.0 0.0 0:00.00 migration_CPU3 6 root 15 0 0 0 0 S 0.0 0.0 105:58.96 keventd
On Fri, 6 Feb 2004 16:40:05 -0900 "Bedard, Joe" <Joe.Bedard@asc.asrc.com> wrote:
Greetings:
Kswapd on my SLES 8 whipsaws one of my CPU's to 99.9% for prolonged periods of time during high I/O. While this activity is happening, performance and interactive response time goes by way of the swamp.
Can you do a kernel profile and post the results? Boot with profile=1 (before triggering the problem) readprofile -r (afterwards) readprofile -m /boot/System.map | sort -n Also could you test if it goes away when you boot with numa=off ? -Andi
participants (2)
-
Andi Kleen
-
Bedard, Joe