suggesting a suggestion.

[xs1]

A very wise comment came to us from another forum, which got me thinking.

i still don't think the speed hickups, or should i say chokes, are a routing problem. i think it's the 100 mbit link to the backbone maxing out. 33 users at 3 mbit=max.

SO, why not limit the tests to seconds?? Sorta like dslreports does. When theirs 50 users testing, it will say

" Test busy, try again in 30 sec "

I think this may limit the bottlenecking, and improve the accuracy since its not trying to burst a milion conections at once.  no?

[RTB]

People will be scared away by that limit, and as you may know, not many like to wait. So unless the routing issues can get fixed without resorting to extreme measures, I'd stay away from such an idea.

[xs1]

People dont like to wait?!  well im sorry but, Fuck 'um. I mean for this site to live you need the users, but at what cost? Testmy wouldnt be accurate anymore, but "testmy.net ; almost accurate" . Their is over 200 users here daily, on this ownage FREE testing site,  0 registering, the ones who DO register cant even take the time to validate the email, and the result ??  People getting innacurate test results.. "Drastic" mesures would be making people register to use the tests. But then they would register, never post and test freely..  This is just a minor adjustment to insure this "accurate" test site becomes once again, just that. Accurate  Because when in the early mornings with noone testing someone can get 10 Mbps... then as they day goes on more users begin testing, and their results become 2 Mbps....thats not sticking to the testmy word.   

[resopalrabotnick]

this would probably entail a whole and complete rewrite of the whole testing script. but what the hell.

reserve 95mbit of the 100 mbit for the test, maybe 15 of that again for the upload.

let the test run as long as there is at least 10 mbit available for the dl test and 2 for the upload. that way many small timers can test simultaneous, only if several fast ones are dl/ing will the script have to delay a few seconds before it starts the queued clients. the values would have to be monitored and tweaked, but 10mbit avail for down sounds like a good value to ensure that even a fast link like RTB's can stretch it's legs.

that would also prevent getting 30 second waits, since the fast links do even the 3 meg test in 10 seconds or less, freeing the bandwidth up quickly. maybe limit number of tests per ip to 5 a day or so or maybe 30 in a week, so someone doing a lot of tweaking can run a hoopload of tests and then just has to wait a bit for his quota to fill up again.

[VanBuren]

the priority nr 1 must be to make testmy.net generate stable accurate speeds, if its a routing issue, lack of bandwidth, or something else, thats CA3LE

[resopalrabotnick]

from the faq:

We do have a fast server - Dual 2.8GHz Xeon Processors, 2 GB DDR RAM, Rehat Linux, MySQL server, Apache Webserver, PHP, 100 Mbps uplink to a 12 Gbps Direct Fiber network.  TestMy.net knows what kind of server is needed to get correct test scores to all users regardless of time of day and server load.

and i don't think, otherwise there is a lot of people with slow connects not taking up much bandwidth, that the number of users shown includes all those people reading/posting on the forum.

only when someone throws down a challenge do they all go to test at the same time, and then get borked results. for my views on adding a short delay, dynamic, at the beginning of the test if necessary, see previous post.

[xs1]

from the faq:

[edit]

and im sorry if others dont agree, but  mr resopalrabotnick  im with you 100 % on this suggestion.

[resopalrabotnick]

it ain't the server that's the problem. it's the network linkup.

wonder CA3le could have them drop a giggabit card in, but then i kind of doubt the site operator would be thrilled at the idea of blasting 1/12th of his total bandwidth with /gasp/ essentially useless speedtest data.

[VanBuren]

from the faq:

and i don't think, otherwise there is a lot of people with slow connects not taking up much bandwidth, that the number of users shown includes all those people reading/posting on the forum.

only when someone throws down a challenge do they all go to test at the same time, and then get borked results. for my views on adding a short delay, dynamic, at the beginning of the test if necessary, see previous post.

If hes limited to the 100Mbps uplink im suprised scores is not even worce.......  Today alot of ppl shooting at 5 Mbps so a average speed should be around 3-4 Mbps

and you can see in my previous post what kinda bandwidth is required for that 

VanBuren

[resopalrabotnick]

not really all that surprising. i dl at about 2800 kbps. even the 3 meg test takes me less than 11 seconds to complete.

that's why the server side should monitor all running tests and delay the start of a test for a few seconds if there happens to be a spike. most people would probably not even reall notice a delay of a few seconds, and the advantage to queing them like this would be that everyone is guaranteed a fat chunk of bandwidth for his/her test.

the accuracy would be much improved as a result, making sure test results are reproducable and the unknown factor of <was the link bottlenecked when i tested or did my tweaking bork mine> would be eliminated.

i for one would be more than happy to wait a few seconds for other testers to finish so i could run a clean test. and that would then be an even better thing to advertise with. testmy.net has this awesome loadbalancing so even though it's the most fucking popular site on the planet everyone gets a fair shot at testing.

[xs1]

hmm wow...  190 users now testing and ...mabey its a fluke... but i actualy got a good score !!!

:::.. Download Stats ..:::

Connection is:: 6225 Kbps about 6.2 Mbps (tested with 2992 KB)

Download Speed is:: 760 KB/s

Tested From:: http://www.testmy.net/

Bottom Line:: 111 times faster than 56K you can download 1MB in 1.35 second(s)

Validation Link:: https://testmy.net/cgi-bin/get.cgi?Test_ID=T3G9LH723

/randomness...  go ahead and keep debating guys, sorry ^__x;

[xs1]

i for one would be more than happy to wait a few seconds for other testers to finish so i could run a clean test. and that would then be an even better thing to advertise with. testmy.net has this awesome loadbalancing so even though it's the most fucking popular site on the planet everyone gets a fair shot at testing.

Lmfao!!  :haha:  

Theirs the testmy quote of 2005!

[resopalrabotnick]

not 190 testing. 190 on site. meaning probably:

100 idling.

50 drooling over their score.

10 cursing their isp for their score

25 perusing the forum

5 testing

proof:

you weren't testing, you were formulating your post, yet you still showed up as on site.

[xs1]

"20 Guests, 8 Users"

Im one of the 28 actualy "on the site" so i dont think i count Theirs at least 100 users testing lol. They dont go to  the testmy homepage and say "omg which should i click  now?! O_O "

[Shug7272]

:huh: Well I think CA3LE should put a orgasm button on the site so I can click it over and over while watching you guys argue over this.

[resopalrabotnick]

so my 25 users on the forum was pretty much accurate.

my point is:

user goes to testmy.net

knocks counter up by one.

klicks on ul or dl test.

selects size of test.

up to now about 15 seconds of essentially idle time have passed. average. give or take for newb / reg.

test starts. i'll give it 15 seconds of actual data transfer. note that the higher the bandwidth used, the shorter the time of transfer!

results show.

user looks them over, ponders them etc.

30 seconds.

maybe a retest. 15 seconds data.

compare results.

cheer/curse

brag to buddies on im or phone or next to him or whatever.

at least a minute

window stays open for reference for further action keeping counter up.

so 2 minutes of a user being active on the site entails 30 seconds of data transfer. ill be generous and let them retest etc, give them 10 megabytes of transferred data which at average speeds will take about 60 to 90 seconds.

and that is at 100 to 1500 kbps, meaning they don't take up much bandwidth.

it's random spikes of users that kill individual tests.

that's what load balancing would eliminate.

[resopalrabotnick]

frag it.

CA3LE, we need usage stats to settle this, lol.

[CA3LE]

frag it.

CA3LE, we need usage stats to settle this, lol.

You are absolutely correct... and this is also why I have changed the verbage of the counter from "XXX users testing" to "XXX users on site" because every page that the counter loads on is counted in the data.  Also, like at the moment it says "221 users on site" - that basically means that in the past 15 minutes (what I have flush set to) 221 unique people have opened some page on the site.  I do have pretty high stats for this site (trending well over 200,000 people this month, an this is growing every month, and one day I will need another test server for sure)... but I don't have high enough stats to have 200 people testing at the same time most of the time... when that counter gets to 2000+ maybe....

As you can see below, (now this is an average... bursts are not shown) I am not using all my resources... as you can see below that, you may see that 60Mbps spike, that is where I was hacked and my box was sending out DOS attacks... my point is that durring that attack it sent 60Mbps+ outbound... no problem.

I am watching the server, if I see this to be a growing issue I may relocate the server.  But right now, alot of the time I test and I get awesome scores (does anyone else?)

:::.. Download Stats ..:::

Connection is:: 5788 Kbps about 5.8 Mbps (tested with 2992 KB)

Download Speed is:: 707 KB/s

Tested From:: http://www.testmy.net/

Bottom Line:: 103 times faster than 56K you can download 1MB in 1.45 second(s)

Validation Link:: https://testmy.net/cgi-bin/get.cgi?Test_ID=LDJ7MU7KE

:::.. Upload Stats ..:::

Connection is: 745 Kbps about 0.7 Mbps (tested with 1013 KB)

Upload Speed is: 91 KB/s

Tested From:: http://www.testmy.net/

Bottom Line: 13 times faster than 56K you can upload 1MB in 11.25 second(s)

Validation Link :: https://testmy.net/cgi-bin/get.cgi?Test_ID=GXUF5XLYO

and from .s1 server

  MySpeed Results 

Download 5,865,688 bps

Upload 812,488 bps

QOS 92%

RTT 62 ms

MaxPause 47 ms

(that upload score is definitely off... not by too much though)

.s1 I am going to PM you some files that I would like you to place on your server so that I can test with my script speed off your server.

If anyone has a fast server and wants to donate bandwidth, please PM me, I am always open for a affiliation.

[resopalrabotnick]

hmm.

the fact that these are averages is the problem.

each sample there is taken over what, 10 or 15 minutes?

if this is indeed a problem with the links bandwidth we are looking for spike with a duration of a few seconds. that is enough to bork a test if someone happens to start in that window. do you have a possibility to monitor your usage with more resolution? or realtime?

the machine you are running sounds big enough to power a 100mbit link fully, so i really am not thinking of that. if anyone out there has an idea if it could be a problem with the config, i would be happy to be wrong about the bandwidth. but to me it sounds like the most probable reason.

worst case: the net the server hooks into has a high load and there isn't a chance in hell that he can hit 100mbps.

only way to check that would be to install the fix to the test i proposed and throttle it way down, until there are no more complaints.also, do you have stats on how many are really testing at any one time?

if there is indeed a problem with the test in any way, finding it will be a major league pain in the arse, lol.

am i glad all i need to do is kibitz.

[CA3LE]

I will setup some deeper stats on JUSt the tests to see how many people are testing at once.  I will let you guys know, we will nail this puppy down. 

[resopalrabotnick]

hm. no, i think you will nail it down. we're just gonna sit here and complain that a certain /free/ speed test is inaccurate.

[reno]

I might be able to help

[resopalrabotnick]

reno, i /was/ kidding. then again, there isn't much we /can/ do, since all the data we'd need to wade through is not yet available to us, and we wouldn't be getting the raw feed, we would be getting data already processed by CA3LE, so all we /can/ do is nitpick.

[reno]

I was talking about when CABLE said "If anyone has a fast server and wants to donate bandwidth, please PM me, I am always open for a affiliation."

Since I have a site and a shit load of rollover bandwidth he might be able to use some. since i'm not right now.

[resopalrabotnick]

hehe

nitpick mode on. but if you host the test part of the month, until your quota for the month is full, wouldn't that make the test even more inaccurate than it already is (or is it?), what with people testing to one site and then another?