kern/73209: PS/2 keyboard problems when NFS server under load (fwd)

Jason Bacon bacon at smithers.neuro.mcw.edu
Tue Nov 2 12:50:32 PST 2004


The following reply was made to PR kern/73209; it has been noted by GNATS.

From: Jason Bacon <bacon at smithers.neuro.mcw.edu>
To: Brooks Davis <brooks at one-eyed-alien.net>
Cc: Brooks Davis <brooks at FreeBSD.org>,
	freebsd-gnats-submit at FreeBSD.org
Subject: Re: kern/73209: PS/2 keyboard problems when NFS server under load
 (fwd)
Date: Tue, 2 Nov 2004 14:42:44 -0600 (CST)

 On Tue, 2 Nov 2004, Brooks Davis wrote:
 
 > On Mon, Nov 01, 2004 at 06:01:28PM -0600, Jason Bacon wrote:
 >> 
 >> Actually, no - the USB disk happened to be connected during boot, but was
 >> never touched.  The process was writing to da0:
 >> 
 >> Filesystem   1K-blocks      Used    Avail Capacity  Mounted on
 >> /dev/da0s1a     253678     56780   176604    24%    /
 >> devfs                1         1        0   100%    /dev
 >> /dev/da0s1e     253678       332   233052     0%    /tmp
 >> /dev/da0s1f   15592330  13594272   750672    95%    /usr
 >> /dev/da0s1d     253678     30288   203096    13%    /var
 >> 
 >> #The following examples export /usr to 3 machines named after ducks,
 >> #/usr/src and /usr/ports read-only to machines named after trouble makers
 >> #/home and all directories under it to machines named after dead rock
 >> stars
 >> #and, /a to a network of privileged machines allowed to write on it as
 >> root.
 >> #/usr                   huey louie dewie
 >> #/usr/src /usr/obj -ro  calvin hobbes
 >> #/home   -alldirs       janice jimmy frank
 >> #/a      -maproot=0  -network 10.0.1.0 -mask 255.255.248.0
 >> #
 >> # You should replace these lines with your actual exported filesystems.
 >> # Note that BSDs export synatx is host-centric vs. Suns "FS-centric" one.
 >> 
 >> /usr/home       -maproot=0      neelix
 >> 
 >> There was nothing else loading the system at the time, either.
 > 
 > That's the sort of information you need to provide in your PRs.  Could
 > you please verify that you can't trigger this behavior with large local
 > accesses?  We need to narrow the field a bit.  Also, please CC   Could
 > you please verify that you can't trigger this behavior with large local
 > accesses?  We need to narrow the field a bit.
 > 
 > Please CC freebsd-gnats-submit at freebsd.org with your reply so this
 > discusion ends up in the PR history.
 > 
 > Thanks,
 > Brooks
 
 Here you go, plus more...
 
 =========================================================================
 Local access on smithers
 =========================================================================
 smithers bacon ~ 218: tar cf x11.tar /usr/X11R6/
 tar: Removing leading '/' from member names
 
 [top on smithers]
 
 last pid: 38077;  load averages:  1.00,  3.21,  3.51    up 6+04:24:28  13:51:38
 61 processes:  2 running, 59 sleeping
 CPU states: 10.8% user,  0.0% nice, 37.8% system,  1.9% interrupt, 49.4% idle
 Mem: 107M Active, 57M Inact, 68M Wired, 11M Cache, 34M Buf, 648K Free
 Swap: 1024M Total, 5776K Used, 1018M Free
 
   PID USERNAME PRI NICE   SIZE    RES STATE    TIME   WCPU    CPU COMMAND
 38077 bacon     -8    0  1624K  1188K biord    0:07 29.49% 20.17% bsdtar
 37868 root      97    0 18500K 17544K select   0:22  2.73%  2.73% Xorg
 37912 bacon     97    0 26488K 16772K RUN      0:20  2.34%  2.34% kdeinit
   502 root      96    0  1268K    68K select   3:32  0.68%  0.68% moused
 37928 bacon     96    0 32388K 17916K select   0:06  0.49%  0.49% kdeinit
 37972 bacon     96    0  2796K  2052K select   0:01  0.15%  0.15% rxvt
 
 [Add another tar while the first is still running]
 
 last pid: 38088;  load averages:  0.93,  2.15,  3.01    up 6+04:27:12  13:54:22
 63 processes:  2 running, 61 sleeping
 CPU states: 17.1% user,  0.0% nice, 41.1% system,  1.6% interrupt, 40.3% idle
 Mem: 108M Active, 65M Inact, 56M Wired, 13M Cache, 34M Buf, 648K Free
 Swap: 1024M Total, 5820K Used, 1018M Free
 
   PID USERNAME PRI NICE   SIZE    RES STATE    TIME   WCPU    CPU COMMAND
 38085 bacon     -8    0  1624K  1192K biord    0:09 12.07% 11.47% bsdtar
 38086 bacon    101    0  1624K  1192K RUN      0:08 12.26% 11.47% bsdtar
 37912 bacon     96    0 26488K 16740K select   0:26  2.83%  2.83% kdeinit
 37868 root      97    0 18664K 17720K select   0:28  2.64%  2.64% Xorg
   502 root      96    0  1268K    68K select   3:33  0.49%  0.49% moused
 37968 bacon     96    0  2488K  1708K select   0:00  0.24%  0.24% rxvt
 38088 bacon     96    0  2376K  1548K RUN      0:00  1.35%  0.24% top
 
 [In nedit on smithers, during 1 tar operation:]
 
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 
 [In nedit on smithers, during 2 simultaneous tar operations:]
 
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 
 No problems with local access.
 
 Repeat the write experiment from a Linux client (neelix):
 
 ========================================================================
 NFS access from SuSE 8.2 client - cp -Rp /usr/X11R6 /smithers/home/bacon
 ========================================================================
 
 [From neelix fstab:]
 
 smithers:/usr/home      /smithers/home  nfs             rw,intr 0 0
 
 [From df on neelix:]
 
 Filesystem           1K-blocks      Used Available Use% Mounted on
 /dev/hda3             73848544  65386420   8462124  89% /
 shmfs                  1034604         0   1034604   0% /dev/shm
 uhura:/home2/jdrive/share
                       75685352  62228256   7402268  90% /windows/share
 smithers:/usr/home    15592330  10064846   4280098  71% /smithers/home
 
 neelix bacon ~ 204: /bin/cp -Rp /usr/X11R6 /smithers/home/bacon
 
 [Top on smithers]
 
 last pid: 38110;  load averages:  2.74,  1.94,  2.68    up 6+04:30:52  13:58:02
 62 processes:  4 running, 58 sleeping
 CPU states:  5.0% user,  0.0% nice, 56.4% system, 13.2% interrupt, 25.4% idle
 Mem: 108M Active, 51M Inact, 68M Wired, 15M Cache, 34M Buf, 648K Free
 Swap: 1024M Total, 5832K Used, 1018M Free
 
   PID USERNAME PRI NICE   SIZE    RES STATE    TIME   WCPU    CPU COMMAND
   359 root       4    0  1256K    12K -       18:05 22.75% 22.75% nfsd
   361 root       4    0  1256K    12K -        8:14  9.33%  9.33% nfsd
   362 root       4    0  1256K    12K -        6:51  8.40%  8.40% nfsd
   360 root       4    0  1256K    12K -        5:47  5.81%  5.81% nfsd
 37912 bacon     98    0 26488K 16740K select   0:34  2.34%  2.34% kdeinit
 37868 root      98    0 18680K 17720K RUN      0:35  1.95%  1.95% Xorg
 
 [In nedit on smithers, during cp from neelix:]
 
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalaalalalalalalalalalalalallalalalalalallalalalalalalalalalalalalalal
 lalalalalalalalalalalalalalaallaalllalalalalalalalalalalalalalalalalalal
 
 (Note the inconsistencies - I was being very careful to alternate between
 'a' and 'l')
 
 
 
 ========================================================================
 Thought of one more angle: See if a FreeBSD client produces the same
 problem:
 
 NFS access from FreeBSD 5.2.1 client (moe) - similar to SuSE client
 (fast Athlon machine)
 
 cp -Rp /usr/X11R6 /smithers/home/bacon
 ========================================================================
 
 [From fstab on moe:]
 
 smithers:/usr/home      /smithers/home  nfs     rw,intr         0       0
 
 [df on moe:]
 
 Filesystem         1K-blocks      Used    Avail Capacity  Mounted on
 /dev/ad0s1a           253678     64400   168984    28%    /
 devfs                      1         1        0   100%    /dev
 /dev/ad0s1e           253678      2556   230828     1%    /tmp
 /dev/ad0s1f        113508718  84828108 19599914    81%    /usr
 /dev/ad0s1d           253678     45282   188102    19%    /var
 apu:/home2         282261358 245038456 14641994    94%    /apu/home2
 apu:/home3         282261358 241584110 18096340    93%    /apu/home3
 smithers:/usr/home  15592330  10064852  4280092    70%    /smithers/home
 
 [top on smithers]
 
 last pid: 38154;  load averages:  0.75,  0.76,  1.45    up 6+04:42:41  14:09:51
 64 processes:  4 running, 60 sleeping
 CPU states:  7.3% user,  0.0% nice, 59.6% system, 15.0% interrupt, 18.1% idle
 Mem: 110M Active, 74M Inact, 50M Wired, 8980K Cache, 34M Buf, 648K Free
 Swap: 1024M Total, 5776K Used, 1018M Free
 
   PID USERNAME PRI NICE   SIZE    RES STATE    TIME   WCPU    CPU COMMAND
   359 root       4    0  1256K    12K -       19:25 32.28% 32.28% nfsd
 37912 bacon     96    0 26488K 16732K select   1:01  2.98%  2.98% kdeinit
 37868 root      97    0 18680K 17704K select   0:53  1.90%  1.90% Xorg
 38078 bacon     96    0  7160K  4396K select   0:10  0.83%  0.83% nedit
   502 root      96    0  1268K    68K select   3:38  0.15%  0.15% moused
 
 * Interesting note: This copy did NOT trigger multiple nfsd processes as
 the Linux client did.  FYI, I've had problems with Linux clients in the 
 past causing panics on my FreeBSD 4.x servers.  I worked around it by 
 using TCP instead of UDP.  ( Followup below )
 
 [In nedit on smithers, during cp from moe:]
 
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 lalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalalala
 
 The keyboard problem does NOT show up when writing from a FreeBSD client!
 
 
 ========================================================================
 SuSE 8.2 client with TCP
 ========================================================================
 
 I then changed the Linux client to use TCP instead of UDP.  SAME PROBLEM
 with the keyboard on smithers, plus the response time on smithers became
 unbearable.  ( The mouse cursor and keystrokes ran several seconds behind )
 
 [From fstab on neelix:]
 
 smithers:/usr/home      /smithers/home  nfs             rw,tcp,intr 0 0
 
 [top on smithers:]
 
 last pid: 38208;  load averages:  5.83,  2.82,  2.12    up 6+05:02:43 
 14:29:53
 64 processes:  8 running, 56 sleeping
 CPU states:  1.6% user,  0.0% nice, 70.3% system, 28.0% interrupt,  0.1% 
 idle
 Mem: 111M Active, 53M Inact, 68M Wired, 9876K Cache, 34M Buf, 648K Free
 Swap: 1024M Total, 5748K Used, 1018M Free
 
    PID USERNAME PRI NICE   SIZE    RES STATE    TIME   WCPU    CPU COMMAND
    359 root      -8    0  1256K    12K biord   22:19 16.21% 16.21% nfsd
    361 root       4    0  1256K    12K -        8:48 13.33% 13.33% nfsd
    362 root       4    0  1256K    12K -        7:17 12.35% 12.35% nfsd
    360 root       4    0  1256K    12K -        6:09 11.52% 11.52% nfsd
 37912 bacon     96    0 26488K 16728K RUN      1:42  0.44%  0.44% kdeinit
 37868 root      96    0 18704K 17756K select   1:27  0.34%  0.34% Xorg
 
 -----------------------------------------------------------------------
 Real men aren't afraid to eat quiche.
 -----------------------------------------------------------------------
 


More information about the freebsd-bugs mailing list