Yet Another Benchmark Script (YABS) - Linux Benchmarking Script using fio, iperf, & Geekbench

Mason · September 14, 2020, 2:35pm

Hi all, please consider participating in the poll below to help me gauge interest in which iperf3 locations should be used by YABS for network testing. There are 18 options, but please narrow down your selection to your TOP 5 choices.

If there’s any reason you DON’T want a specific location added, please voice your concerns here (i.e. if you’ve experienced reliability issues, location is significantly slower than others in the same region, etc.). I appreciate all the help! Feel free to share the poll link with anyone else that regularly uses the script. @Clouvider has recently added public iperf3 servers to all 6 of his POPs, so there’s plenty of “new” locations to consider.

In other news…

YABS updates in the works:

Fix web scrape to get GB4 results from curl (GB’s web design changed slightly, so it doesn’t grab the scores currently) – DONE (implemented in ‘testing’ branch)
Add longer wait time after GB test concludes to allow results to be posted in time – DONE (implemented in ‘testing’ branch)
Add ability to use a GB paid license key to unlock the full GB test – DONE (implemented in ‘testing’ branch)
Update pre-compiled binary for fio (v3.17 → v3.23)
Update pre-compiled binary for iperf3 (v3.7 → v3.9)
Alter iperf3 test locations for the network speed tests based on feedback provided in the poll

Poli · September 14, 2020, 2:54pm

Holy cat @Clouvider is the main sponsor! But would definitely love to see Zurich, to be honest. So I would have a lot of fun testing our few nodes at 10gb/s.

beagle · September 14, 2020, 3:08pm

Since there are more options on Western Europe and US, may I suggest we keep the more “exotic” locations: Bogor, Sao Paulo, Kiev and Moscow independently of the poll results?

Mason · September 14, 2020, 3:43pm

Sounds logical. Poll results won’t be the final say on what the script gets, I’ll keep it diversified

Mason · September 16, 2020, 4:34am

Thoughts on this batch of iperf tests? Interested in what you think about the location choices. Trying to find the right balance of geo-diversity and popular datahubs.

iperf3 Network Speed Tests (IPv4):
---------------------------------
Provider                  | Location (Link)           | Send Speed      | Recv Speed
                          |                           |                 |
Clouvider                 | London, UK (10G)          | 1.97 Gbits/sec  | 2.38 Gbits/sec
Online.net                | Paris, FR (10G)           | 1.87 Gbits/sec  | 2.14 Gbits/sec
WorldStream               | The Netherlands (10G)     | 2.17 Gbits/sec  | 1.19 Gbits/sec
Wifx                      | Zurich, CH (10G)          | 1.70 Gbits/sec  | 488 Mbits/sec
Biznet                    | Jakarta, Indonesia (1G)   | 6.24 Mbits/sec  | 102 Mbits/sec
Clouvider                 | NYC, NY, US (10G)         | 9.39 Gbits/sec  | 9.33 Gbits/sec
Velocity Online           | Tallahassee, FL, US (10G) | 1.84 Gbits/sec  | 2.89 Gbits/sec
Clouvider                 | Los Angeles, CA, US (10G) | 2.37 Gbits/sec  | 2.46 Gbits/sec
Iveloz Telecom            | Sao Paulo, BR (2G)        | 129 Mbits/sec   | 218 Mbits/sec

iperf3 Network Speed Tests (IPv6):
---------------------------------
Provider                  | Location (Link)           | Send Speed      | Recv Speed
                          |                           |                 |
Clouvider                 | London, UK (10G)          | 1.93 Gbits/sec  | 2.35 Gbits/sec
Online.net                | Paris, FR (10G)           | 2.54 Gbits/sec  | 2.16 Gbits/sec
WorldStream               | The Netherlands (10G)     | 2.21 Gbits/sec  | 1.37 Gbits/sec
Wifx                      | Zurich, CH (10G)          | 186 Mbits/sec   | 606 Mbits/sec
Clouvider                 | NYC, NY, US (10G)         | 9.27 Gbits/sec  | 9.28 Gbits/sec
Clouvider                 | Los Angeles, CA, US (10G) | 2.35 Gbits/sec  | 2.46 Gbits/sec

Poli · September 16, 2020, 7:18am

Looks cool to me, would be nice to see some Chinese nodes too now that we are on it, but I don’t know if it’s possible too.

Mason · September 16, 2020, 12:44pm

Really just depends on the iperf servers that are publicly available. APAC doesn’t have any available as far as I’m aware besides the Biznet location. I’ll be on the lookout for more, though

Neoon · October 13, 2020, 10:44am

Architecture not supported by YABS.

Mason · October 13, 2020, 2:01pm

Ooof. What’chu runnin’? BSD?

Neoon · October 13, 2020, 2:31pm

arm arm64 with Ubuntu ok.

root · October 13, 2020, 2:58pm

But while you’re at it, make it run for FreeBSD too

Edit: It also needs to run on dirt-cheap bare-bone offers, like NAT vps where space is extremely small and ram is very low.

imok · October 13, 2020, 3:26pm

It does not work on ZFS either.

Mason · October 13, 2020, 6:15pm

Ahh yeah… I don’t think the fio, iperf, or geekbench binaries will work with arm. Maybe it’s worth exploring making it arm-compatible if I can compile fio and iperf for arm.

I’ll look into it, but I’m not really a BSD guy so no idea what would be involved. But the project is open source for a reason!

Works fine on zfs-backed systems (…if your machine can handle it lol). Fio, iperf, geekbench will naturally try and max out the resource it’s testing and doesn’t really care what else your machine is running. The disk calculation hasn’t been worked out yet since df reports strange values for zfs systems (hence why other bench scripts report more storage space than you actually have).

Neoon · October 13, 2020, 6:24pm

fio and iperf/3 are available on arm sir.
I guess you could at least add a update so it runs without geekbench until you made or not made that working. Nearly everything is available on arm64 but not on armhf which 32bit RPi4’s use.

The 64bit image for RPi4 uses arm64 so no issue

imok · October 13, 2020, 7:12pm

Weird. Tried on 2 Proxmox nodes and both said some kind of error.

imok · October 24, 2020, 4:54am

# ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
#              Yet-Another-Bench-Script              #
#                     v2020-09-21                    #
# https://github.com/masonr/yet-another-bench-script #
# ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #

Fri 23 Oct 2020 11:52:33 PM CDT

Basic System Information:
---------------------------------
Processor  : Intel(R) Xeon(R) CPU E3-1270 v3 @ 3.50GHz
CPU cores  : 8 @ 3785.741 MHz
AES-NI     : ✔ Enabled
VM-x/AMD-V : ✔ Enabled
RAM        : 31Gi
Swap       : 31Gi
Disk       : 437G

Preparing system for disk tests...chmod: cannot access './2020-10-23T23_52_36-05_00/disk/fio': No such file or directory
fio Disk Speed Tests (Mixed R/W 50/50):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       |                   () |                   ()
Write      |                   () |                   ()
Total      | 0.00 KB/s        (0) | 0.00 KB/s        (0)
           |                      |                     
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       |                   () |                   ()
Write      |                   () |                   ()
Total      | 0.00 KB/s        (0) | 0.00 KB/s        (0)
chmod: cannot access './2020-10-23T23_52_36-05_00/iperf/iperf3': No such file or directory

Performing Geekbench 5 benchmark test... *cue elevator music*

Mason · October 24, 2020, 2:36pm

Interesting… And you have read/write access to the current directory? Maybe the curl failed when retrieving the binary file

imok · October 24, 2020, 2:59pm

Tested in 2 servers, same result.

flips · December 6, 2020, 9:35pm

Hm, getting this on a couple of nodes (both NAT nodes and IPv4 public nodes):
Geekbench releases can only be downloaded over IPv4. FTP the Geekbench files and run manually.

On nodes which have both IPv4 and IPv6 … Anything I should check on my end?
(I guess @Mason would be the one to ask here.)

Mason · December 6, 2020, 11:19pm

Means the IPv4 check is failing for some reason. It’s done via a curl to icanhazip.com. Maybe I need to re-evaluate how it determines IPv4 connectivity and switch it to using ping or something like that instead. Unless someone has any better ideas?