Coventry proof of concept trials

[roughbeast]

These trial are due to come to an end by 31st. March. Most of us have been on connections up to 200Mb. The trial, along with the Ashford trial, proved that 200Mb is achievable on ordinary public networks given suitable user equipment. I am now free to give further details about this.

I was luck enough to receive an enhanced modem aimed at much higher speeds, bonding 8 channels down and 2 out of a possible 4 channels up. My best result below was the highest achievable on the set up, but I am reliably informed that the proof of concept trial could have raised the bar even higher. The cryptic name of the trial is merely a device to stop people thinking that VM was aiming at getting 400Mb/40Mb. This was a trial of technical possibilities. I can tell you that during tests my speed dial needle often hit well over 400Mb before it settled down. Needless-to-say such niggles as buffering or lag sounded like fiction on this kind of connection. To have 1Gb downloading in a blink of an eye is a sight to behold. I tried streaming at blue ray quality on 5 machines at one go, whilst using bittorrent on one of those machines. Result? Smooth as a baby's bottom.

For those of you interested I was using a pre-configured modem supplied for the trial. My own WNDR3700 router, capable of almost 500Mb WAN/LAN throughput, and my quadcore processor helped me get consistent speeds over 350Mb. To prevent overburdening our processors whilst testing speeds most trialists uninstalled their regular virus software and installed Microsoft Security Essentials instead. This uses minimal processor time, and it's free. Trialists also applied the TCP optimiser at http://www.speedguide.net/downloads.php, configuring it for connections over 100Mb. This works for Vista and W7.

Sorry I can't give you any more detail than this, but VM will now be confident that 400Mb/40Mb is a viable concept on the existing fibre to the cabinet infrastructure. Why one would need 400Mb download is a completely different matter. :rolleyes: A big household with lots of different users would be the most happy customers I guess, +speed geeks like me.

Last Result:
Download Speed: 371133 kbps (46391.6 KB/sec transfer rate)
Upload Speed: 24064 kbps (3008 KB/sec transfer rate)
Latency: 5 ms
05 February 2011 21:10:08

Virgin Media Proof of Concept Trials (If you're not a registered trialist you'll just get an error message!)

Please note that VMspeed.com is now only accessible to trialists using a trial modem. This was to free it up for trialists only after it went wild a few months ago.

[craigj2k12]

what speedtest did you use to acheive those speeds?

---------- Post added at 22:08 ---------- Previous post was at 22:07 ----------

sorry.... didnt read vmspeeds doh!!!!

---------- Post added at 22:08 ---------- Previous post was at 22:08 ----------

what was the trial modem? not a superhub?

[Nopanic]

:cool:

[qasdfdsaq]

400mb would require me to upgrade my dedicated server to gigabit to properly take advantage of o_0

Interesting info, thanks for sharing, however I do fail to see VM's reasoning and objective for such a trial. Everybody already knows 4-8DS bonding on DOCSIS 3 works, and I don't see any reason to expect it wouldn't. However 8DS is not a practical deployment for a 400/40 consumer service so I don't see the point of testing it that way. Maybe VM think it's good for 200/20, though I personally think it's borderline even for that. I'd have no confidence about 400/40 being viable based on those results, but then again I'm not VM overselling pile high cheap-ass drivel as "service" :p:

I'd also be wary of the "speed dial needle often hit well over 400Mb before it settled down" - I'd ignore this as it's not really indicative of anything. 8 DS channel bonding cannot even theoretically deliver over 400mbps of practical bandwidth, and will not burst higher nomatter how hard they try.

---------- Post added at 22:23 ---------- Previous post was at 22:21 ----------


what was the trial modem? not a superhub?
Lol, a superhub would probably have melted at those speeds :D

[Ignitionnet]

8 downstreams is quite practical for 400Mbps so long as you don't have too many modems sharing - Cablevision in the US sell 101Mbps on 3 US DOCSIS downstreams one of which is shared with 30Mbps customers.

Delivering 100Mbps+ some contention is going to happen, doesn't change the maximum burst speed :)

[craigj2k12]

as has been discussed before, trial users get dedicated lines, more proof of this is "latency: 5ms", do you get this from your line? no.

why do vm run their trialls on dedicated lines, it acheives nothing. vm say ooooh, it works, we will roll it out immediatley. they roll it out on a 20:1 contention and 8 channels between 40 million people they get 2mb each LOL (exaggeration btw)

i thought the promotion with the superhub was "futureproof, up to 400mb"

i want to trial vm products, they give out sweet new modems, and 400meg dedicated lines. it would seem that this "special modem" is virgins way of saying, damn the superhub was a waste, lets make something that works, we will use another modem company than netgear

[TheDon]

as has been discussed before, trial users get dedicated lines, more proof of this is "latency: 5ms", do you get this from your line? no.
Actually, Yes.

https://www.cableforum.co.uk/images/local/2011/03/57.png

[Ignitionnet]

They're proof of concept not trials with a view to production deployments Craig. Just done to make sure hardware performs on live network as advertised as the lab is somewhat different.

Of course these guys do indeed sit on dedicated channels.

---------- Post added at 22:42 ---------- Previous post was at 22:41 ----------

Actually, Yes.

https://www.cableforum.co.uk/images/local/2011/03/57.png

Try something more reliable than a flash app.

Even being the only customer on the channels as I have been I didn't get 5ms.

[TheDon]

Try something more reliable than a flash app.

Even being the only customer on the channels as I have been I didn't get 5ms.

Sure:

Pinging multiplay.co.uk [85.236.96.68] with 32 bytes of data:
Reply from 85.236.96.68: bytes=32 time=7ms TTL=53
Reply from 85.236.96.68: bytes=32 time=6ms TTL=53
Reply from 85.236.96.68: bytes=32 time=6ms TTL=53
Reply from 85.236.96.68: bytes=32 time=9ms TTL=53

Ping statistics for 85.236.96.68:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 6ms, Maximum = 9ms, Average = 7ms

Not quite 5, but still pretty damn low.

Maybe I'm extremely lucky but I've always had stupidly low pings in games, I'm usually always the lowest ping on the server.

[Jayster]

That 5ms is only to the headend though.

[craigj2k12]

i get about 10ms ping to manchester when doing a speed test but 25 when doing a pingtest

[qasdfdsaq]

8 downstreams is quite practical for 400Mbps so long as you don't have too many modems sharing - Cablevision in the US sell 101Mbps on 3 US DOCSIS downstreams one of which is shared with 30Mbps customers.

Delivering 100Mbps+ some contention is going to happen, doesn't change the maximum burst speed :)
Thing is 400mbps over 8 channels gives a single user access to the entire capacity of the bonded set, so as much as one single user downloading at full speed will result in everyone else not being able to hit advertised speeds.

50mb into 200mb while low, does require at least 5 people before the same effect happens - and yes contention does happen but at the same contention ratios, the statistical likelyhood of visible contention is much lower when the pipe is four times the maximum individual user's speed. With a pipe ratio of 1.0, visible contention goes from somewhat likely to an absolute certainty, it becomes impossible to deliver full speed to more than one (!) person at a time on a particular segment of network.

I'm sure you know all the theory already - but do you really think (up to) 100 400mbps users onto a single 400mbps pipe is really reasonable? Sure it's technically possible, but I would be far more comfortable with 16 or 32 downstream channels, to match current ratios (which seem bad enough already in some areas).

I dunno what the service quality is like on Cablevision in the US, but given VM's history I really hold very little hope of VM's network capacity planning to do things well enough to actually make 400mbps work reasonably on a 400mbps pipe.

[craigj2k12]

50mb into 200mb while low, does require at least 5 people before the same effect happens

200/50 = 4 ?

[qasdfdsaq]

That 5ms is only to the headend though.
If only I could get 5ms to my headend...

Pinging cpc9-sgyl31-2-0-gw.sgyl.cable.virginmedia.com [77.99.218.1] with 32 bytes of data:
Reply from 77.99.218.1: bytes=32 time=54ms TTL=63
Reply from 77.99.218.1: bytes=32 time=14ms TTL=63
Reply from 77.99.218.1: bytes=32 time=8ms TTL=63
Reply from 77.99.218.1: bytes=32 time=13ms TTL=63

---------- Post added at 23:02 ---------- Previous post was at 23:02 ----------

200/50 = 4 ?
Exactly. 4 people can get full speed at the same time without affecting each other. At least 5 would be needed before speed degradation can occur.

[craigj2k12]

If only I could get 5ms to my headend...


yeah, headend, DNS server....... the whole VM network????!!

Microsoft Windows [Version 6.1.7601]
Copyright (c) 2009 Microsoft Corporation. All rights reserved.

C:\Users\Craig>ping 194.168.4.100

Pinging 194.168.4.100 with 32 bytes of data:
Reply from 194.168.4.100: bytes=32 time=40ms TTL=252
Reply from 194.168.4.100: bytes=32 time=11ms TTL=252
Reply from 194.168.4.100: bytes=32 time=9ms TTL=252
Reply from 194.168.4.100: bytes=32 time=26ms TTL=252

Ping statistics for 194.168.4.100:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 9ms, Maximum = 40ms, Average = 21ms

C:\Users\Craig>ping cpc1-know12-2-0-cust385.know.cable.virginmedia.com

Pinging cpc1-know12-2-0-cust385.know.cable.virginmedia.com [92.239.253.130] with
32 bytes of data:
Reply from 92.239.253.130: bytes=32 time=8ms TTL=64
Reply from 92.239.253.130: bytes=32 time<1ms TTL=64
Reply from 92.239.253.130: bytes=32 time<1ms TTL=64
Reply from 92.239.253.130: bytes=32 time<1ms TTL=64

Ping statistics for 92.239.253.130:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 0ms, Maximum = 8ms, Average = 2ms

it would seem that my UBR is better than yours lol :p:

---------- Post added at 23:10 ---------- Previous post was at 23:08 ----------

C:\Users\Craig>ping cpc1-know12-2-0-gw.know.cable.virginmedia.com

Pinging cpc1-know12-2-0-gw.know.cable.virginmedia.com [92.239.252.1] with 32 byt
es of data:
Reply from 92.239.252.1: bytes=32 time=7ms TTL=64
Reply from 92.239.252.1: bytes=32 time=12ms TTL=64
Reply from 92.239.252.1: bytes=32 time=10ms TTL=64
Reply from 92.239.252.1: bytes=32 time=6ms TTL=64

Ping statistics for 92.239.252.1:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 6ms, Maximum = 12ms, Average = 8ms

ping the wrong things but still a better result than yours

[roughbeast]

what speedtest did you use to acheive those speeds?

---------- Post added at 22:08 ---------- Previous post was at 22:07 ----------

sorry.... didnt read vmspeeds doh!!!!

---------- Post added at 22:08 ---------- Previous post was at 22:08 ----------

what was the trial modem? not a superhub?

It was an Arris.

[qasdfdsaq]

ping the wrong things but still a better result than yours
Not hard, I'm in one of the worse ex-TW areas.

[roughbeast]

as has been discussed before, trial users get dedicated lines, more proof of this is "latency: 5ms", do you get this from your line? no. why do vm run their trialls on dedicated lines, it acheives nothing. vm say ooooh, it works, we will roll it out immediatley. they roll it out on a 20:1 contention and 8 channels between 40 million people they get 2mb each LOL (exaggeration btw)

I think that VMs strategy is to divide up the available space into as many small slices as possible to have the minimum number of users per slice. This is certainly true of 100Mbs.

[Ignitionnet]

I dunno what the service quality is like on Cablevision in the US, but given VM's history I really hold very little hope of VM's network capacity planning to do things well enough to actually make 400mbps work reasonably on a 400mbps pipe.

http://www.dslreports.com/archive?cid=2

Obviously somewhat tainted, and the service is actually 110Mbit, but you get the idea from all the guys on 30Mbps services on 38Mbps channels.

[Chrysalis]

Thing is 400mbps over 8 channels gives a single user access to the entire capacity of the bonded set, so as much as one single user downloading at full speed will result in everyone else not being able to hit advertised speeds.

50mb into 200mb while low, does require at least 5 people before the same effect happens - and yes contention does happen but at the same contention ratios, the statistical likelyhood of visible contention is much lower when the pipe is four times the maximum individual user's speed. With a pipe ratio of 1.0, visible contention goes from somewhat likely to an absolute certainty, it becomes impossible to deliver full speed to more than one (!) person at a time on a particular segment of network.

I'm sure you know all the theory already - but do you really think (up to) 100 400mbps users onto a single 400mbps pipe is really reasonable? Sure it's technically possible, but I would be far more comfortable with 16 or 32 downstream channels, to match current ratios (which seem bad enough already in some areas).

I dunno what the service quality is like on Cablevision in the US, but given VM's history I really hold very little hope of VM's network capacity planning to do things well enough to actually make 400mbps work reasonably on a 400mbps pipe.

I agree with you strongly here. I dont see whats practical about allowing one single user to burst to their port's entire capacity, its trouble waiting to happen. Personally I think 100mbit is pushing it a bit far on 4 channels as well and that will show problems when it hits heavier areas. I think in reality user's should not be able to burst to more than 25% of given capacity at any time, this allows at least some degree of buffer before congestion occurs. On the 50mbit service assuming 4 50mbit channels this is the case, 25%. Is their a higher mode than qam256 for downstream channels? qam512 maybe?

Of course as well this proof of concept means little in terms of VM network capability as the channels were likely low contention dedicated channels, nothing like what would be what some areas get in a real service. eg. the upgrades done for brighton for the uplifted speeds didnt even remove congestion on the old speeds and the area is now worse off than it was before the changes, this is how far apart trials are from commercial performance.

The evidence is clear.

On 20mbit on non bonded channels I had speed variance from as low as a few mbit up to the full speed, it was all over the place. My burst speed was 40% of the channel's max burst speed. I upgraded to 30mbit, same upstream channel and same downstream channel's but the difference been am now attached to multiple channels at once 200mbit of capacity, so now only 15% of the channel capacity. The difference was instant and obvious, 95% of time able to burst to full speed. Was still some minor fluctuations until my upstream capacity got an upgrade but that indicated most of the downstream congestion was downstream port related and 5% of it or so was upstream related.

---------- Post added at 15:09 ---------- Previous post was at 15:02 ----------

I think that VMs strategy is to divide up the available space into as many small slices as possible to have the minimum number of users per slice. This is certainly true of 100Mbs.

it doesnt matter if its as low as 2:1 contention (which it wont be), contention could still get very obvious.

I think bulldog used to sell some 2mbit adsl over 2mbit virtual pipes, the result was people often seen horrific performance if they were sharing with a heavy user. (which seemed quite common). Or even any user using the connection at the same time.

eg.

20 400mbit users on a 1.6gbit pipe would easily perform better than 5 on a 400mbit pipe, no question about it.

---------- Post added at 15:13 ---------- Previous post was at 15:09 ----------

http://www.dslreports.com/archive?cid=2

Obviously somewhat tainted, and the service is actually 110Mbit, but you get the idea from all the guys on 30Mbps services on 38Mbps channels.

I dont know what their service is like but that sounds crazy to me.

My own connection is clear evidence of how VM wouldnt be able to handle this well.

You know yourself my 20mbit service was poor on a 50mbit channel, yet 30mbit worked fine on 200mbit, and this was the case even before they added the 2nd upstream channel.

[roughbeast]

Why is 200Mb+ not possible as a satisfactory service with multiple users? My understanding is that VM will use 10Gbps pipes. End of problem.

[weesteev]

Why is 200Mb+ not possible as a satisfactory service with multiple users? My understanding is that VM will use 10Gbps pipes. End of problem.

Yes on the Core Network, what the UBR/BSR can handle is a different matter though.