Hello Mikrotik community,
I often use bandwidth test inside ROS to test connection quality about links point to point and customer's adsl.
I always believed this tool is a proper and fast way to test real bandwith avaiable to the user, but now I have many cases where the result do not confirm this likeness between btest and real bandwith.

I did many tests where, from CPE, the btest shows full bandwidth, but in "speedtest.net" the customer obtains about half bandwith.
So I searched the cause in every spot (link wireless, ap, cpe, queues, optical fibers, lan, pc, configurations, traffic monitor, etc...) and I collected a lot of information for a believable statistic.

The first result is when in backhauling there is a PtP wireless link with ccq <60-80%, customers are affected by issues, but they are invisibile to btest.
But now others customers report slowdown:
- btest form cpe to internet -> 100% bandwith
- real bandwith (speedtest, download, iperf same result) is about half.

So the problem is I cannot use btest for real test from cpe to internet, and I must telephone they to see real connection quality. Does exist another way to test bandwidth from remote site without interact with custumers?

My networK
-CPE wireless
-AP nv2/nstreme
-PPPoE authentication
-backhauling: wireless link (nv2)

good idea ......but i think the only way to test that cpe connection with internet is to control it's band traffic ........

Do you test UDP or TCP? while trying speedtest.net etc, you use TCP.

UDP is always better that TCP.

If u use NV2 u may have this kind of problem
I say when u do TCP Btest do it with TCP connection count=1

I have changed all settings in btest but I got the same result.

If u use NV2 u may have this kind of problem
I say when u do TCP Btest do it with TCP connection count=1

That's not a good idea. When you use internet like a normal person, your connection count is always >1. Test 50 connections to get real life test.

I generally see better download speeds than btest shows.

Maybe my CPE config is CPU heavy.

Btest to the CPE? Btest consumes a lot of CPU resources because it generates random data

If u use NV2 u may have this kind of problem
I say when u do TCP Btest do it with TCP connection count=1

That's not a good idea. When you use internet like a normal person, your connection count is always >1. Test 50 connections to get real life test.

Let me explain

I have P2P link witch is about 120Mbit using NV2 -testing with 4 or more connections, when I do Internet speedtest I get around 30mbits
So hire I m confused why is that?
when I do TCP Btest with 1 connection I get around 30-40 mbits (its nature of NV2 thats why we get false throughput)
But when I do simultaneous download of 20-30 files or run bitorent I can get more than 100mbits of download but my Internet speedtest always show 30mbits
like Ciambot wrote: real bandwith (speedtest, download, iperf same result) is about half. -Thats why I m saying test with 1 connection see the result and u can understand why thats happening its NV2

but with nstreme its different story TCP Btest of the link is about 90-95Mbits testing with 1 or 2 connections , when I do speedtest I get around 70-80 mbits

This was explained somewhere in this forum. It's how is and you cannot do anything about it.

My company has a ~200 Mbit fiber connection, I have not once had a "Speedtest" site that could max that connection, even when discounting wireless. We also have a dedicated server hosted in a datacenter about 5 hours away. That server has a 1 Gbit connection and we have run FTP tests that can max our connection.

I can't say exactly what your issue is, however be aware that when you do a speedtest.net or just about any other speedtest site it creates 1 TCP connection for the test. So if your wireless link gets 30-40 Mbits when using btest with 1 stream then it is accurate. Try doing tests to locations that you can control, that are longer then a few seconds, and have more then the needed capacity to saturate your link, and have the ability to create multiple connections.

For us it was approx 10 simultaneous FTP connections in both directions. We test out PtP backhauls in a similar way, and have found that btest is quite accurate for a quick and dirty test, as long as the PC/RB powering RouterOS is fast enough to generate the random date needed.

Hope that helps

Cheers

You are right, friends. Btest shows the real bandwidth available to the customer only with TCP connection count = 1 (or near 1).

Btest to the CPE? Btest consumes a lot of CPU resources because it generates random data

Is this accurate? yo say. TO customer. then the CPE, is not generating packages (TO customer) FROM customer, would be CPU consuming for the CPE. My NAS (pppoe) dont have problems generating packages enough to btest CPE's.

In my test, udp will make more accurate speed test, if CPU is to slow. I have my 18 and 23 Ghz 400mb/s licensed links, and having troble testing bandwith from RB1100AH, due to CPU, but it is 100% accurate at udp test. When RB1100AH is not prosecing packages, just forwarding to next router (MPLS), it have no problem at all giving 400Mb/s (or acurate 371), the same speed tested by udp test.

THis is problem of NV2, there is many topic about it:
http://forum.mikrotik.com/viewtopic.php?f=7&t=66380
http://forum.mikrotik.com/viewtopic.php?f=7&t=66290

If u use NV2 u may have this kind of problem
I say when u do TCP Btest do it with TCP connection count=1

That's not a good idea. When you use internet like a normal person, your connection count is always >1. Test 50 connections to get real life test.

Let me explain

I have P2P link witch is about 120Mbit using NV2 -testing with 4 or more connections, when I do Internet speedtest I get around 30mbits
So hire I m confused why is that?
when I do TCP Btest with 1 connection I get around 30-40 mbits (its nature of NV2 thats why we get false throughput)
But when I do simultaneous download of 20-30 files or run bitorent I can get more than 100mbits of download but my Internet speedtest always show 30mbits
like Ciambot wrote: real bandwith (speedtest, download, iperf same result) is about half. -Thats why I m saying test with 1 connection see the result and u can understand why thats happening its NV2

but with nstreme its different story TCP Btest of the link is about 90-95Mbits testing with 1 or 2 connections , when I do speedtest I get around 70-80 mbits

What cards you uses on nstreme to get 90-95mbit/s?
What routeros version?
Thanks

R52Hn MTv5.21

R52Hn MTv5.21

Dou you have working dual nstreme on N cards?

R52Hn MTv5.21

Dou you have working dual nstreme on N cards?

Never tried only with A-standard cards

R52Hn MTv5.21

Dou you have working dual nstreme on N cards?

Never tried only with A-standard cards

What was your max. throughoutput using A cards? what ros version, what cards, what hardware? thanks

I don't remember quite much it was 2-3 years ago link was about 15km using dishes with dual polarity limb , throughput was about 80 max 90Mbits (TCP one way) using bonding - 2x2 radio cards using 2 different channels in turbo mode
Nstreme Dual (one wlan for tx one for rx) did not work well when running TCP test rx/tx was about 50/10mbit (But 3 years ago on market we don't have hight quality dual polarity antennas like now maybe if we try 2x2 antennas it should go 50/50mbit)
I was using Rb600 and UB5 radio card
when N standard come out instantly I changed with N cards I get about same traffic but using only one 40Mhz channel

Are you using any simple queue?

Are you using any simple queue?

why? silmple queue may cause issues to bandwith test?