Rocketnet - feedback?

[agentrfr]

16:12, now getting 316 KiB/s (that's kibibyte per second, or 2.589 megabit per second)

 

[agentrfr]

I'm going out now, so commentary will continue later

 

[Splash]

Looks like you are downloading pirated movies?

 

[Splash]

I am not sure, but if you believe any ISP will give you 100% dedicated international capacity on FTTH you are dreaming... FTTH is built upon a contention business model, since bandwidth does not cost the ISP the same amount one pays for FTTH.

I did a few tests to a server in USA California and I'm pretty happy with the below based on the fact that international bandwidth is never guaranteed from source to destination since there are too many factors to consider.

I have 2 connections in my home and show the below. Running multiple concurrent connections may produce better results than 1 connection.

SpeedTest servers etc provide an average result based on a quick download/upload. Taking TCP Window sizes in to account, you might find its not long enough to really show whats going on.... hence me using iperf below.

FTTH - 100/100
iperf -c iperf.he.net --parallel 10 -t 30
Cool Ideas [SUM] 0.0-31.2 sec 329 MBytes 88.3 Mbits/sec
RocketNet [SUM] 0.0-31.5 sec 337 MBytes 89.8 Mbits/sec

Cool Ideas:
traceroute to iperf.he.net (216.218.227.10), 30 hops max, 60 byte packets
1 core.splash.co.za (172.29.40.1) 0.195 ms 0.168 ms 0.139 ms
2 u2wc-cust.coolideas.co.za (154.0.14.172) 3.987 ms 3.984 ms 3.949 ms
3 uta-cust.coolideas.co.za (154.0.4.30) 3.930 ms 3.859 ms 3.845 ms
4 unm-cust.coolideas.co.za (154.0.3.82) 4.009 ms 3.993 ms 3.970 ms
5 ur1-cust.coolideas.co.za (154.0.3.205) 3.966 ms 3.911 ms 3.930 ms
6 uod-cust.coolideas.co.za (154.0.3.109) 4.504 ms 4.166 ms 4.117 ms
7 c39-backbone.coolideas.co.za (154.0.2.117) 4.310 ms 4.267 ms 4.212 ms
8 cu-backbone.coolideas.co.za (154.0.2.30) 4.804 ms c5h-backbone.coolideas.co.za (154.0.2.197) 4.823 ms 4.808 ms
9 he.jinx.net.za (196.223.14.108) 166.034 ms 166.016 ms 165.992 ms
10 10ge3-11.core1.lon2.he.net (184.105.81.253) 199.910 ms 199.865 ms 199.825 ms
11 100ge13-2.core1.nyc4.he.net (72.52.92.166) 233.561 ms 240.424 ms 233.519 ms
12 100ge16-1.core1.sjc2.he.net (184.105.81.218) 296.992 ms 296.977 ms 294.615 ms
13 10ge4-4.core1.sjc1.he.net (72.52.92.117) 296.441 ms 100ge13-2.core1.sjc1.he.net (184.105.65.113) 295.765 ms 295.757 ms
14 e0-50.core4.fmt1.he.net (184.105.65.214) 294.578 ms 295.651 ms 295.552 ms
15 216.218.227.10 (216.218.227.10) 295.042 ms 294.999 ms 298.742 ms

RocketNet:
traceroute to iperf.he.net (216.218.227.10), 30 hops max, 60 byte packets
1 core.splash.co.za (172.29.40.1) 0.213 ms 0.188 ms 0.162 ms
2 jhb-io-ter-car-8.directel.co.za (169.255.0.5) 3.473 ms 3.649 ms 3.668 ms
3 jhb-io-ter-ibr-1.directel.co.za (169.255.0.225) 3.979 ms 3.976 ms 3.956 ms
4 he.ixp.joburg (196.60.8.42) 4.023 ms 4.018 ms 4.006 ms
5 10ge3-11.core1.lon2.he.net (184.105.81.253) 172.295 ms 172.296 ms 172.286 ms
6 100ge13-2.core1.nyc4.he.net (72.52.92.166) 233.264 ms 231.741 ms 232.813 ms
7 100ge16-1.core1.sjc2.he.net (184.105.81.218) 294.822 ms 293.172 ms 292.542 ms
8 100ge13-2.core1.sjc1.he.net (184.105.65.113) 294.803 ms 294.440 ms 10ge4-4.core1.sjc1.he.net (72.52.92.117) 293.708 ms
9 e0-50.core4.fmt1.he.net (184.105.65.214) 294.206 ms 296.455 ms 296.399 ms
10 216.218.227.10 (216.218.227.10) 293.118 ms 292.423 ms 294.375 ms

 

[Koosvanwyk]

Latency way down today, about 50% worse than what it was pre 10 Aug 2018. Latency to London about 210ms (compared to 145ms pre 10 Aug 2010), pretty much the same % decline to other international locations ?

Jitter seems a lot better compared to a few days ago


The below graphs represents the average speeds to 5 servers located in the UK with each data point representing 10 test performed between 19:00 and 21:00 at random days during the last 2 weeks (blue line is download and bar graph is upload). The 10th of Aug is where the download speed is 42.1 and everything to the righ of that is after 10 Aug 2018.

What is odd, is that from 10 Aug 2018 the upload speed exceeds the download speed, the slower download speed clear. The last two two tets was performed early in the day (at about 4PM), outside the peak time. From the 10 Aug 2018 the jitter on my line has increased dramatically, until today (I did not test yesterday) as pointed out above, but latency a lot slower.

 

[electrofreeze]

you can almost see the shaping when you look at the straight lines

 

[Splash]

you can almost see the shaping when you look at the straight lines

That doesn't mean you are shaped.....Let me have a chat to the guys at RN, and lets see if we can pin point where the issue is... be it local or with their IPT providers.

Do you have some traceroutes to compare?

 

[Splash]

you can almost see the shaping when you look at the straight lines

RN reported back to me now that their IPT provider has made some changes, please would you do those tests again.

 

[Koosvanwyk]

Seems like it it not just Rocketnet with the slow international speeds and latency all over the place, Coolideas have the same problem, just have a look at that thread.

 

[agentrfr]

Looks like you are downloading pirated movies?

Nah not this time. These are my files I uploaded to my server expressly for this test. I'm doing it this way so that I have a baseline on how their plex server might try stream/transcode directly from the seedbox via an ftps thread

Slightly o.t. but who has time to pirate these days anyway? Netflix is reeeeal good. Between that, apple, hulu and crunchy roll, there is practically no reason to pirate at all. Couple bucks a month and you have access to practically everything being made these days anyway

 

[agentrfr]

I am not sure, but if you believe any ISP will give you 100% dedicated international capacity on FTTH you are dreaming... FTTH is built upon a contention business model, since bandwidth does not cost the ISP the same amount one pays for FTTH.

I did a few tests to a server in USA California and I'm pretty happy with the below based on the fact that international bandwidth is never guaranteed from source to destination since there are too many factors to consider.

I have 2 connections in my home and show the below. Running multiple concurrent connections may produce better results than 1 connection.

SpeedTest servers etc provide an average result based on a quick download/upload. Taking TCP Window sizes in to account, you might find its not long enough to really show whats going on.... hence me using iperf below.
...

I'm well aware of this and I certainly do not expect 100% throughput on a single thread internationally. However, getting 1-3% of advertised line speed for a single thread is not what I would consider to be called fair, right?

 

[agentrfr]

Raw Data Here

https://mega.nz/#F!T90kSS4T!aj0AH5s9V6moXkAwQcC_UA

 

[agentrfr]

So it was pretty painful last night, from about 15:30 until 19:12, speeds for single thread download on an off port were about 3.5mbps (or 3.5% of advertised line speed).

From 19:12 until 21:49, almost a perfect flatline of speed at near as dammit 1mbps.

Then at EXACTLY 21:50, taps are opened slightly and instantly reach about 4.5mbps.

From 21:50 until about midnight, speed more or less on average linearly increases to full speed of 70mbps.

Speed remains at about 70mbps for rest of the night and early morning until my download is finished at about 02:40 AM.

 

[agentrfr]

So here is my gripe:

A single thread download on an off port is perfectly capable of reaching an average of 70mbps when there is enough capacity.

My problem is that it appears to be an artificial rate limiting of download speed (THAT IS SHAPING) when there isnt enough capacity.

My claim is that when shaping is in effect, different destinations/ports are given different priority on the network. For example, a netflex speedtest to netflix's network goes at full line speed at precisely the same time that my other ftps download is being shaped to hell.

 

[iampanga]

So here is my gripe:

A single thread download on an off port is perfectly capable of reaching an average of 70mbps when there is enough capacity.

My problem is that it appears to be an artificial rate limiting of download speed (THAT IS SHAPING) when there isnt enough capacity.

My claim is that when shaping is in effect, different destinations/ports are given different priority on the network. For example, a netflex speedtest to netflix's network goes at full line speed at precisely the same time that my other ftps download is being shaped to hell.

You need to understand that contention occurs at different points in the network path. You are unlikely to contend for bandwidth at infrastructure provider level, slightly more likely at ISP peering level and most likely at ISP transit level. So that may explain why Netflix (which uses a local cache, with free peering with plenty bandwidth) may perform better than an FTP download to a host in EU/USA which relies on an international transit provider.

Best bet, if your current ISP's contention isn't working is to move. It's very unlikely that your ISP is shaping, and more likely that the product you purchased at the price you purchased it is not suited for your use case. Remember the models used by ISPs to determine pricing and usage terms is designed to cater for the "average" individual which may exclude you... @PBCool, have any input?

 

[Splash]

So it was pretty painful last night, from about 15:30 until 19:12, speeds for single thread download on an off port were about 3.5mbps (or 3.5% of advertised line speed).

From 19:12 until 21:49, almost a perfect flatline of speed at near as dammit 1mbps.

Then at EXACTLY 21:50, taps are opened slightly and instantly reach about 4.5mbps.

From 21:50 until about midnight, speed more or less on average linearly increases to full speed of 70mbps.

Speed remains at about 70mbps for rest of the night and early morning until my download is finished at about 02:40 AM.

View attachment 548177
View attachment 548179

Note at EXACTLY 21:50, RN's IPT provider made some changes as I reported.

 

[Splash]

This is the way I see things, so take what I say with a pinch of salt either way.

As a consumer (yes..I'm one) we want to pay as little as possible for as much as possible. If it could be free, even better. ISP's only exist in most cases based on a contented business model and I'm glad they do since we wouldn't have anything otherwise. Long term investments are also made in infrastructure to deliver these services not to mention the really sad exchange rate we face. All/Most ISP hardware is made internationally and with our import charges its crazy what it costs to purchase some decent kit. Eg. A baby Juniper MX80 = +/- R500k, the cost of an average car!

IP Transit is also not as cheap as you would probably expect, and don't forget that you can't run on 1Gbs anymore, its all 10Gb and up which is also costly. So lets take an example.

Lets say average ISP pays anything between R50 to R100 per Mbs of IPT. For this example we will use R60 per Mbs. If you were the only user on the network at 100Mbs, you would be paying the full cost of R6000 for just your non peered bandwidth. This doesn't take in to account any hardware, peering, hosting, support staff, business costs etc. Added to the mix is that the FTTH last mile provider pockets about 70% (thumb sucking) of the cost of your monthly Internet subscription. Working this back, say one pays R1200 for 100Mbs, ISP gets R360 pm. This would, excluding everything else but IPT at an average rate of R60 per Mbs, provide you with 6Mbs of IPT of the available 100Mbs. In reality you would be only paying for about 128kbs or probably less.

In my opinion its pretty unfair to everyone who make use of an ISP's network, especially during peak periods to kick off some downloads and try and max out our your link and then complain about it. Why not do that after hours when most people are no longer on the network and your impact is non existent?

 

[electrofreeze]

This is the way I see things, so take what I say with a pinch of salt either way.

As a consumer (yes..I'm one) we want to pay as little as possible for as much as possible. If it could be free, even better. ISP's only exist in most cases based on a contented business model and I'm glad they do since we wouldn't have anything otherwise. Long term investments are also made in infrastructure to deliver these services not to mention the really sad exchange rate we face. All/Most ISP hardware is made internationally and with our import charges its crazy what it costs to purchase some decent kit. Eg. A baby Juniper MX80 = +/- R500k, the cost of an average car!

IP Transit is also not as cheap as you would probably expect, and don't forget that you can't run on 1Gbs anymore, its all 10Gb and up which is also costly. So lets take an example.

Lets say average ISP pays anything between R50 to R100 per Mbs of IPT. For this example we will use R60 per Mbs. If you were the only user on the network at 100Mbs, you would be paying the full cost of R6000 for just your non peered bandwidth. This doesn't take in to account any hardware, peering, hosting, support staff, business costs etc. Added to the mix is that the FTTH last mile provider pockets about 70% (thumb sucking) of the cost of your monthly Internet subscription. Working this back, say one pays R1200 for 100Mbs, ISP gets R360 pm. This would, excluding everything else but IPT at an average rate of R60 per Mbs, provide you with 6Mbs of IPT of the available 100Mbs. In reality you would be only paying for about 128kbs or probably less.

In my opinion its pretty unfair to everyone who make use of an ISP's network, especially during peak periods to kick off some downloads and try and max out our your link and then complain about it. Why not do that after hours when most people are no longer on the network and your impact is non existent?

lets not forget that there were no problems before the 10th and nobody is trying to max out the line ,even YouTube is buffering ,so the internet becomes pretty useless the bottom line is it was working properly before the 10th and the customer has to tell the isp there is a problem as they either dont care or dont know what is happening on their network they are getting paid to sort this out and that is not happening it is just quiet

 

[agentrfr]

Look, Im well aware of how contention works for ISPs. I dont have a problem with it at all, I really dont.

My issue is that paying R1.4k a month, then on a Friday night get 2-3% of advertised line speed and as such being unable to watch a 240p video stream, is ridiculous.

I dont expect to get 100% line speed 100% of the time. 3% is far off though

 

[Splash]

lets not forget that there were no problems before the 10th and nobody is trying to max out the line ,even YouTube is buffering ,so the internet becomes pretty useless the bottom line is it was working properly before the 10th and the customer has to tell the isp there is a problem as they either dont care or dont know what is happening on their network they are getting paid to sort this out and that is not happening it is just quiet

There are always problems on the Internet, its a dynamic set of networks interconnected, hence InterNet. If your YouTube is buffering, then maybe this is something local to your network? What router do you make use of at home?