CA3LE

ahhh, "My Average" is taking the last 25 results into consideration.  That's a direct copy of how it's currently done.
 
This also filters your results by identifier and host.  So when you're on Spectrum you'll get an average of only those results, when you're on Verizon you'll get an average of your Verizon results.  Same is true if you visit on a desktop, android or iOS.  You'll only compare similar results.
 
But that's only for "My Average" -- the ISP and location are taking into account all results and come from a continuous 15 minute average that runs for all databases.  Those numbers constantly fluctuate with the current average user's performance.

Yes and no... for right now.
 
For download it's enabled by default for upload it only works with multithread... until I push my next update. 
 
You may have already read this topic on it.  I got it fully working just last night.
 
 
As soon as the linear upload test is available in the beta, it will be the default option.  When you toggle the multithread option it will affect both upload and download tests.

sweet.
 
As soon as I posted my response to you on Wednesday I got into building the single thread upload test into the new version.  That inconsistency has been driving me nuts, that it's missing that super important element.  That's what makes TMN different.
 
It's been insanely hard to work that into the new test.  The new version was first built as a multithread test in proof of concept - @xs1 requested a multithread upload test.  Adding the single thread download test to that was simple, bing-bang-boom.  But the single thread upload test is at least 10 times as hard to program.  Even though it's already built... working it into the new program has been a challenge for sure.
 
Anyways, I did it.  It works really good too, graphing the results.  Stripped to the essential functions.  Tested it up to 500 MB... works awesome.  I'll publish the update soon. 
 
 
 
I'm stoked!
  

Hi Rod,
 
I'm happy to see that you're still visiting.
 
I was optimizing the query structure of the database search.  This presented a new problem because now certain calculations are done later in the process.  Not really an issue, we can work with that, especially given the performance gains of dropping 2/3 of the queries... except I can't tell if there should or shouldn't be a graph displayed early enough in the process to generate the page structure the way I wanted.  A problem easily solved if it was a single graph.
 
Instead of just making it work... I rebuilt it into a single graph.  Along the way I finally solved the x-axis / y-axis.  There are two separate y-axis for speed and then two x-axis for date.  The dates align with each other properly now on the x-axis no matter what the situation.
 
 
 
This update also makes DB search query much faster for members like you with over 100,000 results.  My testing saw no less than 2X increase in speed, some queries saw as much as 6X improvement.  Once I thought of how it could be done (early yesterday) it was mostly a matter of reorganizing.  Still a good 20-30 hours of programming, lol.
 
Numerous other bug fixes, more complex queries are possible.  Thank you @Pgoodwin1 for the suggestions, got the juices flowing.
 
Thank you @rrr10 -- sorry it took >5 years to program your suggestion.  It's a harder problem than I ever could have imagined.  No examples to go off of, very little documentation of this very specific use case.  I guess not very many people want to combine charts in this way... well, I like it.  I can see how it will help people understand results better.  I hope you like it too.  

Thanks
 
What a bunch of improvements, you are a rock star...
 
How I can to donwnload a report of all test, with the min and max values? is posible?
 
Thanks

Hi Ca3le
 
I take some reviews using the the multithread and "single" test, and the single test is much better that multithreat, I am thinking on that, becouse have more servers saturating a connection must be the best option, but like satellite options using dynamic assignemt like tdma or fdma for upload and is a shared media looks that could be that Starlink protect itself for multitrhead delivery to protect his backbone.
 
I test in other enviroments using a multiflows froma single source and give better results that using a single one, but I think the winner aproach is as you mention have a warm period, in other solution even delimite the measure time to try to only measure in the peak rate, like out a pause time of 10 second and meause only for 20 seconds.
 
I will continue doing more test at leats for the next 3 weeks.
 
Thanks 

Hi, can I download a report of the speedtest maximum and minimum bw per down and up for each test?, now I can download the average.
 
Thanks

Hi Rod,
 
I'm happy to see that you're still visiting.
 
I was optimizing the query structure of the database search.  This presented a new problem because now certain calculations are done later in the process.  Not really an issue, we can work with that, especially given the performance gains of dropping 2/3 of the queries... except I can't tell if there should or shouldn't be a graph displayed early enough in the process to generate the page structure the way I wanted.  A problem easily solved if it was a single graph.
 
Instead of just making it work... I rebuilt it into a single graph.  Along the way I finally solved the x-axis / y-axis.  There are two separate y-axis for speed and then two x-axis for date.  The dates align with each other properly now on the x-axis no matter what the situation.
 
 
 
This update also makes DB search query much faster for members like you with over 100,000 results.  My testing saw no less than 2X increase in speed, some queries saw as much as 6X improvement.  Once I thought of how it could be done (early yesterday) it was mostly a matter of reorganizing.  Still a good 20-30 hours of programming, lol.
 
Numerous other bug fixes, more complex queries are possible.  Thank you @Pgoodwin1 for the suggestions, got the juices flowing.
 
Thank you @rrr10 -- sorry it took >5 years to program your suggestion.  It's a harder problem than I ever could have imagined.  No examples to go off of, very little documentation of this very specific use case.  I guess not very many people want to combine charts in this way... well, I like it.  I can see how it will help people understand results better.  I hope you like it too.  

Hi,
  When I view the 'My Results' graph, the x-axis is the day and time of the tests. Many times at a particular point on the x-axis there can be a download result at a different day and/or time than an upload result. I would expect that they both would have the same time stamp at a point on the x-axis. Does this make sense? Can this be fixed?
thx
Rod
 

I'm not sure either because I haven't been able to ever get it to do that.
 
My guess is something on the provider end is messing with TestMy.net's output. It may be trying to accelerate previously loaded content.
 
I do a lot on my end to try to bust host caching and clear your cache programatically. But you may get an occasional outlier. Seems to really only affect the multithread download test.
 
You can delete that outlying result by checking the box next to that result and click the trash can at the top of the result details table.
 

 
When I'm not programming or building server infrastructure on the backend I often scan the results for outliers. After so many years of doing this and adjusting the program they're pretty few and far between but I would like to develop a solution for any remaining.
 
When I've rotated back into programming mode I'll make sure to make the necessary adjustments to prevent this. I think before the result is logged your average will be taken into account, it's already being queried at that stage. If the result is much higher than your average and the result is over 1 Gbps a re-test occurs with further cache busting measures. After re-test, if the second result is within a range of the first result it will allow logging. And if not, the test will die with an error so we don't create a loop.
 
When I find these in the results I usually see the client re-test soon after, the re-test always returns to the median.
 
 
It would really help if I could reliably make the issue happen but it's an intermittent issue I've never personally seen. Unless I witness it I can only try to fix it blind. But I think we have a roundabout solution, just need to program it in there.
 
Thank you for the constructive feedback, it's always very helpful. You are building TestMy.net.

It will affect your average but very little given the number of results you have. If you delete it then it won't be averaged.
 
I can delete it for you if you'd like but I thought you might want to try it yourself.

My beta gives anyone the ability to run TMN on any computer they'd like.  Run it locally on a Docker images I've prepared for you or install it on any web sever, with or without SSL.  It only takes 2 files, a total of 1800 bytes of code on your end.  What speed will you see between two wired 1 GbE?   Exactly what you'd expect to see, watch for yourself.
 
tmn-on-my-server.mp4
The Mac and Windows machines are fairly distant, non-direct routes.  The connection traverses 2 switches, then the router, then another switch also adding in about 100 ft of cable before it arrives.  Not lab conditions, I wanted all of those real world variables. 
 
If you inspect I think you'll agree, TMN's results are pretty exact. 
 
tl&dr

 
So average 916 Mbps | 904 Mbps.  Adding 6% network overhead you get to 971 Mbps | 958 Mbps.  Which is right in line with what we see coming across the interface in reality.  
 
I did the same LAN testing when developing the current version (v18) you're using now, always do.  But now I'm giving the ability to do it yourself, super quick.  ... I've also done that before but not like this.  This is very different.  Cut and paste a few commands into Terminal or PowerShell and you're running local tests on all your devices in a few seconds.  You can also audit all of the code and understand the entire container in just a few seconds.
 
Pretty excited to share that with you all.  Curious how people will use it.

As always, thank you for your excellent suggestion.
 
My plan is to basically duplicate what they do over at PassMark's site cpubenchmark.net
 
https://www.cpubenchmark.net/compare/5027vs5684vs5172vs2830/AMD-Ryzen-9-7900X-vs-Intel-i9-14900KF-vs-AMD-Ryzen-5-7600-vs-Intel-Xeon-E5-2667-v4
 


 
Definitely won't be easy.  But comparison like that is extremely helpful.
 
Side note: pretty sure I'm going with the Ryzen 7900X on the new database server.  Currently running Dual E5-2667 v4.  The 7900X hits where I need it most (single thread) and a single CPU will blow the doors off what I'm running right now.  ... couple that with Ryzen's ECC memory support and then they say that the AM5 socket will be the platform until 2027 so upgrade options for many years. -- when you run a mysql database query, that's a single thread operation.  Excited to get it built and compare performance.
 
So imagine that comparison but with internet benchmarks.  Easy to understand and easy to use.

As always, thank you for your excellent suggestion.
 
My plan is to basically duplicate what they do over at PassMark's site cpubenchmark.net
 
https://www.cpubenchmark.net/compare/5027vs5684vs5172vs2830/AMD-Ryzen-9-7900X-vs-Intel-i9-14900KF-vs-AMD-Ryzen-5-7600-vs-Intel-Xeon-E5-2667-v4
 


 
Definitely won't be easy.  But comparison like that is extremely helpful.
 
Side note: pretty sure I'm going with the Ryzen 7900X on the new database server.  Currently running Dual E5-2667 v4.  The 7900X hits where I need it most (single thread) and a single CPU will blow the doors off what I'm running right now.  ... couple that with Ryzen's ECC memory support and then they say that the AM5 socket will be the platform until 2027 so upgrade options for many years. -- when you run a mysql database query, that's a single thread operation.  Excited to get it built and compare performance.
 
So imagine that comparison but with internet benchmarks.  Easy to understand and easy to use.

Much appreciated

Pgoodwin1
Hi CA3LE. Got an odd bug you can maybe fix.
 
I go to DB/Fastest ISPs/Full Listing, I searched on Cincinnati Bell and Spectrum. I pulled up the Speed Test Logs for both - separate windows to compare
 
It defaults to All Identifiers in the Results tables. I set it to 1000 / page. OK.      I select the “Speed” column heading so that it shows the fastest. OK.   Everything’s fine.
 
The issue: In the Results Table, I can’t choose to display the fastest Speeds first in the Speed column when selecting an Identifier (other than All Identifiers)
 
If I select an Identifier (like Dallas, TX), the Speed column automatically reverts to the slowest Speed of the 1000. 

When I select the Speed column heading to make it show fastest first, the Identifier automatically reverts to “All Identifiers”. When I reselect Dallas TX, it automatically displays the slowest Speed first again.
 
haha - an endless cycle of despair

Pgoodwin1
BTW:  My personal test results “My Results” table doesn’t behave that way. You can select an Identifier then sort on the fastest speed without it reverting to All Identifiers

Pgoodwin1
@CA3LE just wondered if you had a chance to look into this. If you don’t plan to, that’s OK too.
 
CA3LE
I'll fix this for you and hit you back.  

Pgoodwin1
😁
Great

CA3LE
Did some updates to the database query.  It should now work the way you were expecting.
 
Let me know if it's querying correctly now.
 

Pgoodwin1
It’s working perfect now. Thanks. That’s a very cool tool.
 
maybe in the future you could add the date range field like our personal results page has. That way you could look at the ISPs performance over a specific period.
 
one suggestion on our date range in our personal results: make it so you don’t have to scroll through the Calendar. When you want the start date to be 3 years prior to today’s date, it takes a long time to get there. Maybe use two boxes, one for start date, one for end date, and let the user be able to type in the date as an option. More error prone that way though, but if it showed the date format, us idiots should be able to get it right.
 
i don’t want much. HAHA🤗

CA3LE
  14 hours ago, Pgoodwin1 said:
maybe in the future
 
The future is now.  You can now search hosts and locations by date range.
 
Currently this pulls data from a table where the old results are purged.  So will be limited to about the last 45K results per host.  Within those results you can now perform more fine grained searches.
 
I'm working on a super server upgrade that will cut down query times on my larger tables.  This should enable us to run queries like never before on the larger datasets.
 
I have what I think is going to be a very interesting topic on the subject.  Where I'm going to detail my build process.  All of my servers are custom built but this new server is going to be my most custom build ever.
 
Basically, found that enterprise grade hardware (that I've been using) isn't the way for TestMy.net moving forward.
 
I was able to beat TestMy.net's current Dual Xeon E5-2667 v4 w/ Optane setup using a single Optane and an old i7 10700K.  Not by a little either.  It seems core performance is the bottleneck.
 
The 13th and 14th gen Intel chips have oxidation and voltage issues so I think I'm going to end up going with AMD.  Also Ryzen supports ECC memory.  I'm basically building a Ryzen gaming PC inside of a 2U rack mount chassis.  Also building a piKVM v2 into it to basically give the machine IPMI.  Be able to remotely see the display, control input, mount ISOs... control it like I'm there.
 
I've got the build coming in right now at less than $1K.  That's about what I paid for just the CPU's and Optane drive in the master server.  And I got them used, 1/6 the original price.  
 
After testing I'm convinced, my dream server is a gaming PC.  Enterprise hardware, even the most recent CPUs... right there with basic consumer CPUs.  They support more memory, have more cache but the performance per core where it counts most is far, far lower.  My concern with hosting TestMy.net is usually focused on single thread performance.  Similar to how I focus on single thread performance in my tests but this is talking about CPU threads.  More threads is definitely helpful and nessesary but the speed of each thread is the most important variable.
 
The single thread performance I'll get out of any AM5 Ryzen will absolutely crush any of the Xeon CPUs in my price range.  The ones I'm looking at out perform or are at par with even the latest enterprise CPUs.  At the end of the day I can build servers cheaper with the latest chips.  All working together in my proxmox cluster so hardware failure is tolerable... but I imagine they'll be just as stable, under volting the CPUs and run them through extra stability testing before production.
 
There's so much competition in the gaming PC performance arena which leads to less expensive consumer chips being faster.
 
And yes, better database structure solves problems.  But so does adding more power.  It's far easier for me to give my cluster more power than to restructure databases.  Things I've built recently and going forward are better planned but there's all the databases and tables prior.
 
If you don't mind, I'd like to share this thread as a topic.

Pgoodwin1
Wow. That sounds really good. It does totally amaze me what you can get in processing power these days. Consumer grade machines can be so fast and not that expensive.
 
“If you don't mind, I'd like to share this thread as a topic.”  Certainly go ahead and share it as a topic
 
thanks for the upgrade with the date range. That tool is awesome now for comparing ISPs. I’ve been looking at maybe changing from Spectrum to AltaFiber (which is Cincinnati Bell). That change you made just adding the date range makes it very easy to compare apples to apples - different test servers, 
 
No problem, thank you for the suggestion.
 
CA3LE
No problem, thank you for the suggestion.
 
The limitation on the host and location queries has needed that correction for a long time.  Once I know someone is trying to query in a broken way, it's motivation to make it work the way you expect.

It may have been a routing issue.  Has it cleared up yet?
 
If I saw something like that happen as @xs1 said, I'd run traceroute and compare to a known working machine.
 
As for getting it fixed.  Depends on where the issue is, could be on your provider's end, my provider's end or one of their peers in between.  Usually, these kinds of problems resolve on their own.
 
It may be that there was only an issue going to and from that specific datacenter.  As long as you're testing good across the greater majority, I wouldn't worry about it too much.  Unless it starts affecting the real communication that you do.
 
Important thing... we know it's not unique or localized to your computer or network.  You've proven this by testing across multiple networks and showing greater results.  

When the auto speed test was first designed in 2010/2011, phones were still figuring things out.
 
Always, my goal with design is to make ONE thing that works with all browsers and devices.  So I've always avoided apps.  TestMy.net has never had an Android of iPhone app.
 
Now it's 2024, phones are a different breed.  They are full fledged gaming PCs that make the desktop computers from 2010 look like a joke.  In 2010 average RAM was 3-4GB.  Now, our phones are coming out with 8-12GB!
 
The philosophy remains, ONE thing for them all.
 
For the auto test to work on any machine, not just Android, make sure the power remains on and the test in focus.  On my S22 I normally have my display set to never turn off.  I'm now noticing that they must have recently updated... now you can't set anything higher than 10 minutes.
 
That's alright, there's an app for that.
 
Screen Alive on Android does the trick.  It's free with an option to donate to the developer.
 
Surprisingly, this isn't an issue on iOS.  I would have thought if anyone did this it would be Apple.  Android usually does things first but not things like this.  On iOS just go do Display & Brightness > Auto Lock > Never.
 
Once your phone's display doesn't sleep anymore you can plug it into power and start up the auto test.  Keep the browser in focus and it will do its thing.
 
Please let me know if this helps

Prefect - thanks for extensive reply - worked great. Focus is key I think, different to PC where I can have browser in background. Obvs phones do it differently in background permissions...
 
 

Id try a traceroute from the affected machine vs one that is working properly. 
 

Running into a problem at a specific location complaining of slow websites (medical/documentation site).
 
Upload speeds for all those other sites show nearly ISP stated speeds (100/100 dedicated fiber).
testmy.net upload for my closest server (Dallas) shows a miserable 300kbps
testmy.net upload to Colorado Springs, CO shows nearly ISP stated speeds (100/100 dedicated fiber).
 
The windows server at this location testmy.net upload to the closest server (Dallas) shows nearly ISP stated speeds (100/100) for locations other than Dallas which shows around 10Mbps (still many many times faster than workstations)..  So the issue is largely limited to the closest Dallas server on the workstations on this domain only.  Has me a bit stumped, any ideas?  
 
Have pointed the workstations direct to 8.8.8.8 on DNS to try and eliminate a DNS problem with the server.  
 
Thanks,

Sorry it's taken me a couple of days, I'm developing.
 
If you toggle the beta in My Settings, then visit the upload or download test there's an explanation.
 

 
There are reasons I originally did it that way and they still hold true.  It makes for a more consistent test with far less variables.
 
Man!  Your suggestion for the upload test back in 2018.  That's an algorithm I still use.  Has saved a lot of bandwidth and time... no joke, over the hundreds of millions of upload tests since, probably something like a decade or more of wait time saved at this point.  I can calculate the bandwidth saved... I'll just estimate quickly, going off the recent upload test results
 
Keep in mind, what's logged to the database is the final result.  The client may have cycled though up to 4 tests before getting to the final test.
 
2 GB in the last 4 minutes (non-peak early morning hours)
 
1440min per day / 4minutes = 360 * 2GB = 720GB per day
 
The optimization was made 6 years ago
 
720*365*6 = 1,576,800 GB or 1.6 Petabytes!
 
Wow.  I think you could easily figure an extra 30+% for the pre-tests.  So about 2 PB saved!
 
Using the current median download speed from the recent download test results of 62 Mbps we can get a rough estimate of the time saved.
 
byte conversion
62/8 = 7.75 MB/s
 
2 PB is 2147483648 MB
(2*1024*1024*1024)
 
2147483648 / 7.75 = 277094664 seconds saved
 
(((277094664 / 60sec per min) / 60min per hr) / 24hrs per day) / 365days per yr = 8.8 YEARS SAVED!
 
The actual number is probably much higher.  Your post, I'd say has saved at least 10 years of wait time collectively.  Amazing.
 
I'm keeping your histogram idea at the forefront of my mind too.    -- I'll post over there.

@CA3LE Further on this, can't you make the upload/download be a single stream that cuts off once enough data has been gathered? I don't understand why the multiple steps are needed.
 
(also, if you could chime in on this I'd greatly appreciate it 🙏)
 

I'm trying to compare iWay and Init7 in Switzerland, and in the rankings iWay seems to score twice as fast as Init7, but when you go to the individual stats, Init7 way outscores iWay.
 
Ranking: https://testmy.net/country/ch#isps
#1 iWay: 1811.4 Mbps 2152 Mbps  316 Mbps   47 clients took 356 tests recently #2 Init7: 624 Mbps 871 Mbps  154 Mbps   67 clients took 183 tests recently  
Individual stats:
iWay: https://testmy.net/hoststats/iway, Averages 326.7 Mbps / 85.3 Mbps | 81 ms, last 1000 download: 195.7, upload: 117.4 Init7: https://testmy.net/hoststats/init7, Averages 621.6 Mbps / 227.8 Mbps | 72 ms, last 1000 download: 1265.3, upload: 146.9

I don't understand how these numbers are related? In any case, I think a histogram would be way more useful for the individual stats, it would show the different kind of connections an ISP has.

Changed it so it looks for /b/imb/blank.png first, then defaults to favicon.