Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Ixel]

I'm sorry but I must also disagree craigski. Someone performing a speed test at 3am should honestly be able to expect to see close to or above 900 megabits. This is a problem that rightfully needs reporting to the ISP and needs to be pushed on until there's a solution. I can understand potentially seeing slower speed during peak time but not in the early hours of the morning when a lot of people are likely asleep. Unfortunately this isn't the first time in the last couple of years that Zen has had this kind of issue reported by some customers. It was pretty much the reason why I moved away from Zen quite a few years ago.

[Alex Atkin UK]

It's a known issue that Zen have acknowledged and are investigating.

The GEA problem sure, but if its there is a single-threaded performance problem on other backhauls, that changes things a bit.

[bogof]

I just want to put to  bed any thoughts that these speed test servers are not somehow capable...

I just hired an Ubuntu instance of a 25G connected Elastic Compute Cloud virtual server on AWS.  Speed testing is close to 10G for download on both the servers I was using to test the Zen link.
The guy I was talking to at Zen today was telling me I shouldn't believe what the Zen server says for speedtests as it isn't very capable... Utter piffle.  I'm beginning to lose the will to talk to these folk.

Code: [Select]

   Speedtest by Ookla

     Server: Zen Internet - London (id = 40788)
        ISP: Amazon.com
    Latency:     1.70 ms   (0.06 ms jitter)
   Download:  9373.96 Mbps (data used: 4.7 GB )
     Upload:  4780.34 Mbps (data used: 2.2 GB )
Packet Loss:     0.0%
 Result URL: https://www.speedtest.net/result/c/a70cd9eb-b2e9-4dc8-a35f-29d85520dcc2

   Speedtest by Ookla

     Server: Swish Fibre - London (id = 34948)
        ISP: Amazon.com
    Latency:     2.67 ms   (0.03 ms jitter)
   Download:  8420.16 Mbps (data used: 7.6 GB )
     Upload:  4776.85 Mbps (data used: 2.3 GB )
Packet Loss:     0.0%
 Result URL: https://www.speedtest.net/result/c/1ae21155-ce84-4b65-bc16-db3a4796fca5


[bogof]

I'm sorry but I must also disagree craigski. Someone performing a speed test at 3am should honestly be able to expect to see close to or above 900 megabits. This is a problem that rightfully needs reporting to the ISP and needs to be pushed on until there's a solution. I can understand potentially seeing slower speed during peak time but not in the early hours of the morning when a lot of people are likely asleep. Unfortunately this isn't the first time in the last couple of years that Zen has had this kind of issue reported by some customers. It was pretty much the reason why I moved away from Zen quite a few years ago.

I think furthermore, given that the line from me to the exchange seems quite capable of the 900, if there are backend issues that make attaining the higher throughput levels problematic, I almost might as well just have a 500 connection - which I'd imagine would seem to be able to be saturated most of the time (single threaded issues not withstanding).

The GEA problem sure, but if its there is a single-threaded performance problem on other backhauls, that changes things a bit.

I've not really looked into the single threaded performance much, but I should I guess.  Unfortunately the command line Ookla speed tester doesn't seem to have a single threaded option, so I'll have to see about trying something perhaps with iperf on that Amazon server.  It only cost me about 25p to play around with a couple of EC2 25G capable servers for a couple of hours yesterday, so I might give that a go this evening again.

[craigski]

It is quite literally sold as having an average of 900Mb/s.
Zen have Samknows boxes in the homes of random, geographically diverse users and they must meet an average of 900Mb/s to be able to advertise the service as it is.

Reading these testing principles, https://samknows.com/blog/testing-principles it seems there is some structure and thought gone into the methodology, they are performed in quiet times, and tests are scheduled so they don't clash. I would expect that would yield more predictable and consistent results than random user initiated tests on speedtest.net.

If the Sam Knows results are giving an average of 900Mbs for a random sample of Zen 900 subscribers, does it suggest there could be an issue with Zen using the speedtest.net testing methodology vs the Sam Knows methodology? It would be interesting to know how those tests differ (Sam Knows vs speedtest) for those random Zen 900 subscribers.

Unfortunately the command line Ookla speed tester doesn't seem to have a single threaded option, so I'll have to see about trying something perhaps with iperf on that Amazon server.  It only cost me about 25p to play around with a couple of EC2 25G capable servers for a couple of hours yesterday, so I might give that a go this evening again.

25p well spent, interesting to compare. I am seriously interested in results as an observer.

[Moderator edited to remove the excess trailing blank lines.]

[craigski]

I did my own experiment. I went to speedtest.net in Chrome browser, I selected 'Zen Internet London', I opened developer tools to try to find out what speedtest.net is doing under the covers, as I'm curious.

It appears to be downloading lots of random 25MB files using port 8080 and timing those downloads. What is interesting is they are not all from Zen IP address space. It looks like its downloading from 4 servers:
51.148.82.21 (this is 5 hops away, Zens server)
5.61.120.37
45.92.46.21 (goes via twelve99.net,  telia.net, 12 hops)
217.138.226.208 (this looks like m247 that is 15 hops away on traceroute)

The dev tools in Chrome will show you the times for each of the 25MB downloads, vastly different depending on what server its downloading from.

[Ixel]

I did my own experiment. I went to speedtest.net in Chrome browser, I selected 'Zen Internet London', I opened developer tools to try to find out what speedtest.net is doing under the covers, as I'm curious.

It appears to be downloading lots of random 25MB files using port 8080 and timing those downloads. What is interesting is they are not all from Zen IP address space. It looks like its downloading from 4 servers:
51.148.82.21 (this is 5 hops away, Zens server)
5.61.120.37
45.92.46.21 (goes via twelve99.net,  telia.net, 12 hops)
217.138.226.208 (this looks like m247 that is 15 hops away on traceroute)

The dev tools in Chrome will show you the times for each of the 25MB downloads, vastly different depending on what server its downloading from.

If it's a multiple connection test then it will potentially choose a few other servers in addition to the chosen one. This can be seen on a result afterwards by going to the results page and then clicking on the '+x more' in the 'Location / Server' column.

For one of my multiple connection speed tests I saw the following:

Code: [Select]

London
Lightning Fibre Ltd

Additional servers used for download test
Eastbourne
CloudConnX

Eastbourne
M-Tech Systems

Maidstone
Trooli

[bogof]

I will try and work out later how to check what the command line tester is doing.

[Alex Atkin UK]

Provisioning confirmed I'm on WBMC (BT Wholesale).

I have speedtest-cli on my router and tried a few different servers it shows as the nearest.  Some of those were really bad (double digit speeds) so certainly if its using some of those the result is going to be WAY off.

Single-threaded tests just now, I've removed upload tests as that always seems to be fine:

Code: [Select]

Hosted by Kubbur (London) [4.20 km]: 29.846 ms
Testing download speed................................................................................
Download: 136.04 Mbit/s

Code: [Select]

Hosted by Zen Internet (London) [5.75 km]: 33.774 ms
Testing download speed................................................................................
Download: 203.08 Mbit/s

Code: [Select]

Hosted by Telxius (London) [5.75 km]: 78.502 ms
Testing download speed................................................................................
Download: 146.17 Mbit/s
I get this is test is during peak hours, but its pretty much what I was getting at 3am too.  Also the tests being quite short aren't giving it time to reach peak speeds as seen from my earlier iperf3 tests.

Worth noting, the Zen server immediately after this became unavailable which perhaps means its overloaded so speedtest.net dynamically removed it from the list?

I will confirm however that multi-threaded speeds are absolutely fine especially real-world.  I actually managed to hit 180MB/s off Steam yesterday as I have Steam servers load balanced across both WANs.  Was getting up to 915Mbit off Zen, 500Mbit off Three 5G.

[craigski]

I have speedtest-cli on my router and tried a few different servers it shows as the nearest.  Some of those were really bad (double digit speeds) so certainly if its using some of those the result is going to be WAY off.

This is my point, the speedtest.net is at best an estimate of your speed. Its random what servers its using, even if you select what you think is a good one, any of the other 3 may not be so good at the specific time you are doing the test.

fast.com, this is similar uses 4 server on the netflix network, but uses 26.2MB files, and port 443/HTTPS

BTw test uses a different approach, a single IPv4 address on akamai, looks like it runs 4 threads to same server downloading 1GB file, for 5 seconds, and calculates how much its downloaded of the 1GB in 5 seconds, again port 443/https

I go back to my original point I just don't trust the numbers on speedtest.net site.

[bogof]

This is my point, the speedtest.net is at best an estimate of your speed. Its random what servers its using, even if you select what you think is a good one, any of the other 3 may not be so good at the specific time you are doing the test.

fast.com, this is similar uses 4 server on the netflix network, but uses 26.2MB files, and port 443/HTTPS

BTw test uses a different approach, a single IPv4 address on akamai, looks like it runs 4 threads to same server downloading 1GB file, for 5 seconds, and calculates how much its downloaded of the 1GB in 5 seconds, again port 443/https

I go back to my original point I just don't trust the numbers on speedtest.net site.

The web based speed tester does spread to multiple servers sometimes.  I don't know what logic it employs to do that - I think it might be based on it detecting an issue between you and the server selected, but it usually tells you about it (or at least it used to).  It would say something like "zen and 3 others"
The command line based speedtest application I can confirm seems to only test to the single server you specify.  I just watched it using iptraf-ng from the command line on my 25G connection virtual server at Amazon, and you can see all the connections are to a single machine at Zen over port 8080.  (speedtest02a.web.zen.net.uk:8080;).  So it would appear that machine is on a 10G capable network.

[bogof]

Worth noting, the Zen server immediately after this became unavailable which perhaps means its overloaded so speedtest.net dynamically removed it from the list?

The Zen and Swish fibre servers both work from speedtest command line application from speedtest.net, but speedtest-cli doesn't work with either. 

[craigski]

If I am reading this correctly, @Alex says his line is performing well, but the speedtest 'estimate' is not consistently giving him good results? Isn't this more weight to my point you can't rely on the speedtest.net results? If its using 1 + 3 random servers, you will need avg ~250 from each, Alex was not getting that, but getting good 'real world' results:

I will confirm however that multi-threaded speeds are absolutely fine especially real-world.

Another theory, I will assume the clever people that have designed the speedtest.net system and also those that have installed the speedtest servers, have built some filtering and restrictions in place to prevent users constantly running speedtests impacting users streaming video, music, making VoIP calls, ie speedtest is considered low priority network traffic, and network designed accordingly.

Maybe this is why speed.cloudflare.com or fast.com give higher 'estimates', as they have more control of the servers on their network, ISP will have a faster connection to those netflix/AWS/cloudflare networks than a random speedtest server at another ISP?

@bogoff has already confirmed that there is more than 1G connection from AWS to Zen speedtest server. I want to see the results from AWS to @bogoff, I happy to pay the 25p 

[craigski]

Here is a simple 2 minute test I have just done to prove the speedtest 'estimate' is wrong in 'estimating' connection speed.

1) Run a speedtest using Zen server to get an 'estimate', if you use Chrome dev tools you can see the servers IP addresses.
2) Immediately after run a test on fast.com whilst monitoring your WAN utilisation on your router WAN interface, this will validate the fast.com/netflix 'estimate'

Obviously taking into account any CPU / PPPoE limitations on your flavour of firewall, you are not maxing our a core etc, so maybe worth monitoring the firewall CPU whist running the test, as monitoring bandwidth will also require some CPU cycles. Also need a reasonable spec PC as test is intensive in chrome browser. But you all know that anyway.

The fast.com server selected should be on the Zen network, so will use the GEA link from exchange direct to Zen network, If I'm understanding correctly what a GEA link is.

netflix-cache1.lond2.ptn.zen.net.uk [51.148.80.10]

Is the result from (2) higher than (1) ?

[skyeci]

EDIT- Zen came back to me again confirming I am on BTW..

I was interested to find out if I was a gea or BTW for my recent new service.. zen report I am neither as I am City fibre apparently which I was not aware of... have they got this right as my upgrade was direct with zen...somewhat confused.