1

I'm trying to host java8 UI application on VPS server (Digital Ocean) with 1GB of RAM. Java is started with -Xms600m -Xmx600m(startup and max memory usage). Also it runs VNC + mate-core. My assumption is that 400m should be enough for non-java staff and 1GB is OK for me.

But I experience strange memory problems. With no swap and swappiness=0 system after some time goes into 90% CPU usage by kswapd0 and system freezes.

I've added swap of 500MB. With same swappiness=0 now top shows that both almost all physical memory is used plus whole swap too. I was trying to check what is consuming so much memory and see in top that sum of RES is around 730M - seems to be as expected.

In my understanding - with such memory usage swap is not needed at all, but actually whole swap of 500М appears to be consumed - so total memory consumption (physical + swap) appears to be close to 1400M and I still experience strange freezes and during these moments kswapd0 is the first one in top.

So ...

  • What am I getting wrong here?
  • Why "Mem Used + Swap Used" is twice higher than "Sum of RES"?
  • Why there is so high Swap usage with swappiness=0?

top sorted down by "RES" (After hint from @TooTee - added columns SWAP and USED, which is RES+SWAP)

top - 14:12:17 up 14:13,  2 users,  load average: 1.40, 3.49, 2.99
Tasks: 115 total,   1 running,  78 sleeping,   0 stopped,   0 zombie
%Cpu(s): 17.6 us,  8.5 sy,  0.0 ni, 65.2 id,  0.2 wa,  0.0 hi,  0.2 si,  8.3 st
KiB Mem :  1009136 total,    66492 free,   881084 used,    61560 buff/cache
KiB Swap:   511996 total,     1056 free,   510940 used.    22580 avail Mem

  PID    VIRT    RES   SWAP   USED    SHR S %CPU %MEM     TIME+ COMMAND
 3271 3065936 690516 274056 964572    400 S 21.8 68.4 212:57.30 java
 1348  298512  46952  78576 125528   6476 S  6.2  4.7  62:43.95 Xvnc4
  883  496100  12328   2968  15296   7688 S  0.0  1.2   0:41.00 do-agent
 3256  633796   9728   7912  17640   6132 S  1.0  1.0   3:47.81 mate-terminal
 1475  544648   5100   5736  10836   3252 S  0.2  0.5   3:37.92 clock-applet
    1   77684   2572    116   2688    532 S  0.0  0.3   0:11.74 systemd
 1476  493176   2492   4580   7072     44 S  0.0  0.2   0:04.09 wnck-applet
 1412  442048   2340   5208   7548    704 S  0.0  0.2   0:06.59 marco
 1400  992492   1916   5924   7840    304 S  0.2  0.2   3:30.01 mate-settings-d
 1100  111460   1828    440   2268      0 S  0.0  0.2   0:00.00 (sd-pam)
 8684   23072   1644      0   1644     24 S  0.0  0.2   0:00.08 bash
  409   80244   1268   1172   2440   1152 S  0.0  0.1   0:08.14 systemd-journal
 8697   44552   1072      0   1072    388 R  0.2  0.1   0:07.14 top
 1434  700512   1024   6876   7900      0 S  0.0  0.1   0:02.43 mate-panel
  703   70756    584    524   1108    404 S  0.0  0.1   0:02.32 systemd-resolve
  865   31748    512    224    736    436 S  0.0  0.1   0:00.89 cron
 8585  107776    464    856   1320    300 S  0.2  0.0   0:00.81 sshd
 1506  323368    408   1416   1824      0 S  0.0  0.0   0:00.04 polkit-mate-aut
  891  287992    376    792   1168    288 S  0.0  0.0   0:04.66 accounts-daemon
 1359  436920    356   7596   7952      0 S  0.0  0.0   0:00.49 mate-session
  887   44640    312    780   1092     60 S  0.2  0.0   1:54.70 openvpn
  892  263036    180   1112   1292      0 S  0.0  0.0   0:03.61 rsyslogd
 1084  107776    176    844   1020      0 S  0.0  0.0   0:00.21 sshd
 1231  107776    160    868   1028      0 S  0.0  0.0   0:10.07 sshd
 1375  220776     72    716    788      0 S  0.0  0.0   0:00.05 at-spi2-registr
  776   43564     48    944    992      0 S  0.0  0.0   0:00.62 systemd-udevd
  678   71972     36    696    732      0 S  0.0  0.0   0:00.53 systemd-network
  869   50188     36    716    752      0 S  0.0  0.0   0:00.40 dbus-daemon
 1358   38572     36    376    412      0 S  0.0  0.0   0:00.10 vncconfig
  860   70580     32    692    724      0 S  0.0  0.0   0:00.37 systemd-logind
  889  170884     32   7816   7848     12 S  0.0  0.0   0:00.15 networkd-dispat
  597  141924     24    576    600      0 S  0.0  0.0   0:00.33 systemd-timesyn
  885   28332     24    192    216      0 S  0.0  0.0   0:00.00 atd
  944   16412     24    124    148      0 S  0.0  0.0   0:00.00 agetty
  925   72296     16    740    756      0 S  0.0  0.0   0:01.69 sshd
  945   14888     16    112    128      0 S  0.0  0.0   0:00.04 agetty
 1092   76772      4   1180   1184      4 S  0.0  0.0   0:00.03 systemd
 1320   63764      4    484    488      4 S  0.0  0.0   0:00.00 sudo
 1366  349216      4    784    788      4 S  0.0  0.0   0:00.00 at-spi-bus-laun
 1449  281988      4    824    828      4 S  0.0  0.0   0:00.01 gvfs-gphoto2-vo
 1460  267396      4    684    688      4 S  0.0  0.0   0:00.01 gvfs-goa-volume
 1465  269200      4    696    700      4 S  0.0  0.0   0:00.01 gvfs-mtp-volume
    2       0      0      0      0      0 S  0.0  0.0   0:00.01 kthreadd
  • Please use the f key in top to enable the SWAP column to see how much swap is each process using. I guess you'll find that the desktop environment is mostly swapped out and/or that a sizable chunk of JVM is on swap as well. You're asking for 600 MiB of heap to be accessible to the Java app, but the JVM also needs memory for its internals, stacks and so on. – TooTea Jul 15 '19 at 12:26
  • I've added columns SWAP and USED and now I see that Java seems to be actually eating 960MB. Am I correct now? If yes, than 1GB of total RAM is most probably not enough for such hosting - because of too much swapping. – Xtra Coder Jul 15 '19 at 14:24
1

I'l try to answer my own questions using the knowledge I've gained so far

What am I getting wrong here?

The major wrong thing is assumption that process' memory usage is reflected in RES column. It is actually reflected in USED column (= RES + SWAP) and top does not show USED column in default configuration (I'm curious why) - type f key to configure needed visible columns, order and sort order.

Another major wrong thing is estimation of memory usage by java process - actual=~960MB vs. expected=~650MB. This means that whole physical address space of 1GB may get consumed by that only java process.

Why "Mem Used + Swap Used" is twice higher than "Sum of RES"?

Because, as I see now from top, some processes take more space in SWAP than in RES and in my particular case that results in ~50/50 distribution of overall memory usage between RES and SWAP

Why there is so high Swap usage with swappiness=0?

Because actual memory usage is indeed high and processes could not fit into physical memory.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.