Speedtest Option

Just now, Stuart Weenig said:

Wouldn't this be similar to just running an external web check in LM, but with more metrics?

The problem with most of the web-only options is the upload speed test is either nonexistent or very inaccurate. The speedtest-cli client is not supported and does not use the same protocols normally used by Ookla tests.  I have not tried the SpeedOfMe API yet, so not sure if it is better or worse, but would hope they account for running it similarly to interactive conditions.  I also just found Ookla now has a supported CLI option, which looks hopeful.  Seems like it needs a license for commercial use, though.  https://www.speedtest.net/apps/cli

2 minutes ago, mnagel said:

I checked again just now for better options and found https://speedof.me/api.html; trying that to see how well it may work. Not free, but not terribly costly either.

Wouldn't this be similar to just running an external web check in LM, but with more metrics?

Yeah,  I have that working, at least from Linux collectors (with puppet ensuring package deployed).  Not ideal and not terribly accurate.  I checked again just now for better options and found https://speedof.me/api.html; trying that to see how well it may work. Not free, but not terribly costly either.

https://pypi.org/project/speedtest-cli/ ?

Looks like Akamai has stopped this option to test a download. Does anyone have any other options?

Had to go with what works... see:

I found some even more promising options:

• https://www.powershellgallery.com/packages/Speedtest/2.0/Content/Speedtest.ps1
• https://github.com/originaluko/SpeedTester

I was able to make the first one work with barely any changes, but the results are not accurate.  The second one needs to be tortured a bit to make it suitable since emits live status data, but again, running it from the CLI it is not accurate.  The only CLI option I have found that is accurate is speedtest-cli (Python).  Looking at both the above to see how hard they would be to rewrite into Groovy and make accurate.

My current iteration of the original code is below (for now), but this method is also not very accurate.  It relies on two properties, one optional:

speedtest.run: set to 1 (or anything) for hosts that should run the code (may instead use manual instances to enable multiple test sources).
speedtest.testmy.mirror: shortname of the testmy.net mirror (e.g., lax, ny, etc.).  if not provided, will use default mirror selection.

$size = 50;
$mirror = "##speedtest.testmy.mirror##";
if ($site -eq "") {
    $url = "https://testmy.net/dl-${size}MB"
}
else {
    $url = "https://${mirror}.testmy.net/dl-${size}MB"
}
$path = "Out-Null"
$WebClient = New-Object System.Net.WebClient 
$mbps = "{0:N2}" -f (($size/(Measure-Command {$Request=Get-Date; $WebClient.DownloadFile( $url, $path )}).TotalSeconds) * 8)
Write-Host "Mbps="$mbps

1 hour ago, mnagel said:

I just found this while trying to see what options we had for this capability.  What I originally hoped to do was to deploy the speedtest CLI package and run in JSON output mode to get proper up/down results (which I may still do at some point).  For the moment this works, if you fix that math error :).  I also changed the applies to expression to "run.speedtest" so I can choose specifically which hosts are involved -- normally would be a site collector, but might not if a site is only monitored remotely.  To get similar location flexibility with the speedtest CLI method would be tricky.

I misread the code -- it does not use PSRemoting, so it can only run from a collector.  Will have to see if I can fix that :).

I just found this while trying to see what options we had for this capability.  What I originally hoped to do was to deploy the speedtest CLI package and run in JSON output mode to get proper up/down results (which I may still do at some point).  For the moment this works, if you fix that math error :).  I also changed the applies to expression to "run.speedtest" so I can choose specifically which hosts are involved -- normally would be a site collector, but might not if a site is only monitored remotely.  To get similar location flexibility with the speedtest CLI method would be tricky.

One thing to note about datasource YTFWWC  is the calculation, which seems to be wrong.

It downloads a 50MB, then it divides how long it took to download (in seconds) by 100 and then it multiplies it by 8 (to convert from MB to mbps)

I think this division should be 50 instead of 100.