[Bug 253968] 13.0-BETA4: UFS with journaled soft-updates is three times SLOWER than before

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=253968

--- Comment #6 from Mark Millard <marklmi26-fbsd at yahoo.com> ---
(In reply to Mark Millard from comment #5)

[I should have mentioned that the ThreadRipper
results are based on non-debug builds of main
(14-CURRENT). So are the below. Same FreeBSD
vintage (same source tree even). UFS2+SU (no J)
again as well.]

aarch64 examples that others might be able to
compare against . . .

>From a OverDrive 1000 (4 Cortex-A57 cores
8 GiByte RAM):

# time -l make extract
===>   firefox-85.0.2,2 depends on file: /usr/local/sbin/pkg - found
===> Fetching all distfiles required by firefox-85.0.2,2 for building
===>  Extracting for firefox-85.0.2,2
=> SHA256 Checksum OK for firefox-85.0.2.source.tar.xz.
      221.36 real        57.32 user        65.94 sys
     76068  maximum resident set size
        42  average shared memory size
        12  average unshared data size
       128  average unshared stack size
     32646  page reclaims
        16  page faults
         0  swaps
     22893  block input operations
   1175983  block output operations
        47  messages sent
        45  messages received
         5  signals received
   1198799  voluntary context switches
       487  involuntary context switches

# time -l rm -fr /wrkdirs/usr/ports/www
      111.95 real         0.48 user        18.95 sys
      3756  maximum resident set size
         7  average shared memory size
         3  average unshared data size
       126  average unshared stack size
       596  page reclaims
         0  page faults
         0  swaps
         2  block input operations
    555695  block output operations
         0  messages sent
         0  messages received
         0  signals received
    613033  voluntary context switches
         1  involuntary context switches

The "rm -fr" here took a larger fraction of the
"make extract" "real" time compared to the
ThreadRipper ratio.

A MACCHIATObin Double Shot got similar results
(4 Cortex-A72 cores, 16 GiByte RAM):

# time -l make extract
===>   firefox-85.0.2,2 depends on file: /usr/local/sbin/pkg - found
===> Fetching all distfiles required by firefox-85.0.2,2 for building
===>  Extracting for firefox-85.0.2,2
=> SHA256 Checksum OK for firefox-85.0.2.source.tar.xz.
      209.41 real        49.36 user        62.94 sys
     75620  maximum resident set size
        42  average shared memory size
        12  average unshared data size
       127  average unshared stack size
     33507  page reclaims
        10  page faults
         0  swaps
     13861  block input operations
   1176990  block output operations
        47  messages sent
        45  messages received
         5  signals received
   1189697  voluntary context switches
       293  involuntary context switches

# time -l rm -fr /wrkdirs/usr/ports/www
       99.39 real         0.48 user        19.33 sys
      3752  maximum resident set size
         7  average shared memory size
         3  average unshared data size
       124  average unshared stack size
       596  page reclaims
         0  page faults
         0  swaps
         2  block input operations
    555695  block output operations
         0  messages sent
         0  messages received
         0  signals received
    639878  voluntary context switches
        29  involuntary context switches

Of course, the SSD's involved are not likely to
be as subject to fragmentation consequences as
spinning drives. So comparisons to the above
might be complicated by such.

(There may be better things to look at with
other tools.)

-- 
You are receiving this mail because:
You are the assignee for the bug.