n-st / nench Goto Github PK

The CPU benchmarking values measured using time are not redirected to the file appropriately, since time prints to stderr by default. I.e. when running bash nensh.sh > nensh.txt the times will be printed to the console (stderr) and not show up in nensh.txt

Changing the lines

    time "$gnu_dd" if=/dev/zero bs=1M count=500 2> /dev/null | \
        "$@" > /dev/null

to

    ( time "$gnu_dd" if=/dev/zero bs=1M count=500 2> /dev/null | \
        "$@" > /dev/null ) 2>&1

Fixes the issue. There may be an easier solution to this, I'm not well versed in shell programming.

"column" binary not standard on Ubuntu, need to install package bsdmainutils

dd test is displaying error -

dd test
nench.sh: line 161: printf: 2.3e+09: invalid number
    1st run:    0.00 MiB/s
nench.sh: line 164: printf: 2.4e+09: invalid number
    2nd run:    0.00 MiB/s
nench.sh: line 167: printf: 2.6e+09: invalid number
    3rd run:    0.00 MiB/s
nench.sh: line 171: printf: 2.43333e+09: invalid number
    average:    0.00 MiB/s

I should mention this is on a ZFS based fs.

The server hosting wget.racing is currently (and likely permanently) offline. A replacement is being prepared, apologies for the inconvenience.

Ubuntu 20.04

Disks:
sda   38.2G  HDD

CPU: SHA256-hashing 500 MB
    CPU: bzip2-compressing 500 MB
    CPU: AES-encrypting 500 MB
    
ioping: seek rate
    min/avg/max/mdev = 102.7 us / 315.6 us / 19.9 ms / 252.1 us
ioping: sequential read speed
    generated 7.51 k requests in 5.00 s, 1.83 GiB, 1.50 k iops, 375.6 MiB/s

While in CentOS 7, 8

Disks:
sda  152.6G  HDD
sdb    100G  HDD

CPU: SHA256-hashing 500 MB
    2.062 seconds
CPU: bzip2-compressing 500 MB
    6.318 seconds
CPU: AES-encrypting 500 MB
    1.390 seconds

ioping: seek rate
    min/avg/max/mdev = 93.3 us / 363.1 us / 17.9 ms / 363.8 us
ioping: sequential read speed
    generated 5.03 k requests in 5.00 s, 1.23 GiB, 1.01 k iops, 251.7 MiB/s

Connection to speedtest.dal01.softlayer.com over IPv6 fails.

--2019-07-19 14:22:21--  http://speedtest.dal01.softlayer.com/
Resolving speedtest.dal01.softlayer.com (speedtest.dal01.softlayer.com)... 2607:f0d0:1003:31::2
Connecting to speedtest.dal01.softlayer.com (speedtest.dal01.softlayer.com)|2607:f0d0:1003:31::2|:80... failed: Resource temporarily unavailable.
Retrying.

is it okay if i change the test to a another DC with working IPv6?

• speedtest.dal05.softlayer.com
• speedtest.dal06.softlayer.com
• speedtest.dal09.softlayer.com

Downloading executable code over HTTP is probably not the best idea.

One can download from https://github.com/n-st/nench/raw/master/nench.sh instead, but the script also downloads ioping binary over HTTP.

The static binary was originally added for three reasons:

• Less effort and system clutter:
  Among nench's software requirements, ioping is arguably the most uncommon
  and least likely to be present by default.
  Providing the static binary avoids additional installations, at least on
  binary-compatible systems (Linux amd64 in this case).

• Consistent output format:
  ioping's output format changed between version 0.9 and 1.0, but 1.0 still
  reports the old version:

  % apt-cache policy ioping
  ioping:
    Installed: 1.0-2
    Candidate: 1.0-2
    Version table:
   *** 1.0-2 500
          500 http://httpredir.debian.org/debian stretch/main amd64 Packages
          100 /var/lib/dpkg/status
  % ioping -v
  ioping 0.9
  

  The same happens with the static binary, but there I can be certain which
  version it is (and parse the output appropriately).

• Consistent test scope:
  ioping 1.0 added a second summary line (marked with > below) that reports
  iops and throughput for the full deadline period (-w) requested on the
  commandline, rather than the old summary that excludes data from some of the
  probes (such as the "warmup" at the beginning):

  % ioping -DRL -w 5 . # ioping 0.9
  
  --- . ( ) ioping statistics ---
  301 requests completed in 5.11 s, 58 iops, 14.7 MiB/s
  min/avg/max/mdev = 778 us / 17.0 ms / 236.0 ms / 31.4 ms
  
  % ioping -DRL -w 5 . # ioping 1.0
  
  --- . ( ) ioping statistics ---
  1.65 k requests completed in 4.99 s, 413.5 MiB read, 331 iops, 82.9 MiB/s
  > generated 1.66 k requests in 5.00 s, 413.8 MiB, 330 iops, 82.7 MiB/s
  min/avg/max/mdev = 2.75 ms / 3.01 ms / 66.0 ms / 2.10 ms
  

Long story short, the static binary could be made superfluous by switching to
the "opportunistic" summary line (which is present in both old and new
versions) and selecting it by its content rather than its position in the
output.

It would also be possible to use the -B option (which I wasn't aware of when
I created nench, but which is present even in older versions) and parse the
"raw" output it produces. It may be possible to re-use the Bps_to_MiBps
formatter that's already used for the dd and curl output.

The primary disadvantages would be potentially breaking comparability between
old and new nench results, and requiring the installation of ioping just
for the test.
Perhaps offer the static binary for manual installation (then documented in
the README) and use only the system-installed version if the user doesn't
request otherwise?