50Mbit Rollout Has Started

[popper]

LOL , 2Gbps, 10Gbps, 20Gbps and 50Gbps indeed....

http://www.techwatch.co.uk/2008/05/1...l/#comment-389
May 13, 2008
Virgin Media completes high speed trial

Originally Posted by broadbandbug

There is no kit that VM haven't bought that enables Upstream Bonding.. The Cards for Upstream Bonding are not yet available from either Cisco or Motorola.. 2009/2010 as far as I am aware.
The only option in the meantime is more dense modulation schemes (QAM32 or QAM64) and you know how that would perform

Err...but they used upstream bonding when the trial first started.

They had 6mbit upload (1.5mbit per channel) at the start of the trials, it was only recently (well 5+ months) that they changed it down to 1.5mbit.

[spiderplant]

Originally Posted by TraxData

Direct from the horses mouth about 50mbit, apparently they cant do simple math.

Please can you run us through the simple math? I think it may be too simple.

I did a quick test with a 20Mbps modem earlier. Doing a single FTP, I achieved a 46:1 downstream:upstream ratio, and that was without any tweaking of the receiving system.

If you Google for "TCP highly asymmetric", you'll find lots of papers on the subject. This is one of the more readable.

When links are highly asymmetric (say more than about 48:1) there is simply not enough “return” bandwidth to keep TCP sending at full speed.

[foddy]

Originally Posted by spiderplant

Please can you run us through the simple math? I think it may be too simple.

I did a quick test with a 20Mbps modem earlier. Doing a single FTP, I achieved a 46:1 downstream:upstream ratio, and that was without any tweaking of the receiving system.

That DOES sound a bit simple ... at what layer were you measuring this bandwidth? From 1:46, I'd assume you're just measuring IP packets (40/1500 would be 1.15:46).

But, you probably performed this test over ethernet. On top of your IP packet, you have an ethernet frame plus additional protocol overheads. This would normally add up to 38 bytes (ethernet addresses, packet length, checksum, guard space, etc.)

38 bytes on top of a 1500 byte data packet is 2.5% extra. 38 bytes on top of a 40 byte ack packet is 95% extra, almost double.

So, for ethernet your measured ratio would be 1.95:49.025, assuming everything above; that's close enough to 2:50.

Now, there are two BIG caveats, and one little one.

Firstly, the cable modem is NOT ethernet. I really don't know enough about it to know what sort of overheads there are, but you can be sure there's a source and destination address. The fact that the cable side has MAC addresses which look just like ethernet addresses would suggest that they could be comparable.

Secondly, I've no idea at which protocol level the bandwidth limit is at. If it's discarding the lower-layer protocols before calculating the speed, then 1.33Mbps would be sufficient (although NOT enough to you to make best use of the bandwidth).

If the bandwidth limit INCLUDES the additional overheads (whatever they may be) then it's unlikely that 1.33Mbps would be enough: even 12 bytes (two mac addresses) is enough to increase the bandwidth requirement to 1.77M.

Thirdly, delayed-ack could reduce the ack traffic by half. It will for some people ... but from your calculations above, not you.

Of course, this is all academic if you want to do anything except downloading large files. Most people will be want do to things like check for e-mails, send e-mails, read web pages, have Windows doing updates, NTP updates, weather updates, RSS feeds.

More upload would be beneficial, but then again it's more than you'd get with 20Mbps so it's a move forward.

[broadbandbug]

Originally Posted by popper

this lack of tested upstream bonding "silver" or "Full" D3 seems to be true at the moment, however 2009/2010 seems far to long a commercial availablility timescale given the new push to the new (Euro-)D3 spec worldwide.

while its not really a viable option as far as i can tell,(does anyone have experience of this kit?) they could go for a limited "Full" D3 Speced "Casa Systems" device or 3 in each headend as a limited trial/test on the "Full" D3 certified spec, if matched up with one of the "Full" D3 CPE (probably Ambit on past experience)due out in full production in around 3 months i think....

also,there are several more "waves" to come this year, so there is also a potential to get a silver or even a full 'Cisco or Motorola' card through and into commercial production this year....

http://www.lightreading.com/document...53441&site=cdn
"CableLabs also awarded “full” qualification status for Cable Modem Termination Systems (CMTS) to Casa Systems for two devices. This is the first qualification of a DOCSIS 3.0 CMTS representing “full” or complete compliance with all requirements of the DOCSIS 3.0 specification for headends. Motorola received bronze qualification for its CMTS."

---------- Post added at 19:50 ---------- Previous post was at 19:16 ----------



i was under the impression the No.s your quoting are the average US non Euro D1.0 No.s

and the majority of kit in VM network was infact Euro D1.1 kit ? and so has both more freq and speed to play with on average....

and thats not even allowing for the far improved speeds/freq per single channel of the new (Euro-)Docsis2.0B spec that the (Euro-)D3 will be using as its base....

also , i cant find the link i put up on the other thread that showed the recommended "(US)industry best practice" slides, was to in effect re-seg/shuffle the available total currently used freqs around, so as to allow far for scope in improved larger blocks of freqs to play with later,to get ready for the up and coming D3 rollouts worldwide....and other future options etc.

sure its on a company by company basis, but it seems only wise of VM with its massive UK market share, to have at least started , or at least got a plan ready for the recommended re-shuffle of the total used freq plans over the whole VM network.

and thats not even considering the turning off of the old analogue sometime in the future (2010/1012 was it?).

The majority (Two Thirds) of VM Network is US DoCSIS, only the Bromley Platform is EuroDoCSIS. For Upstream Bandwidth they are both the same with regards to channel width and capacity availability etc.

---------- Post added at 21:54 ---------- Previous post was at 21:50 ----------

Originally Posted by TraxData

Err...but they used upstream bonding when the trial first started.

They had 6mbit upload (1.5mbit per channel) at the start of the trials, it was only recently (well 5+ months) that they changed it down to 1.5mbit.

Propriatary non standard CMTS and Modems from Arris.. As you know they didn't win through the vendor selection (too expensive?).
Motorola and Cisco do not have an upstream channel bonding capability currently available.

---------- Post added at 21:56 ---------- Previous post was at 21:54 ----------

Originally Posted by foddy

That DOES sound a bit simple ... at what layer were you measuring this bandwidth? From 1:46, I'd assume you're just measuring IP packets (40/1500 would be 1.15:46).

But, you probably performed this test over ethernet. On top of your IP packet, you have an ethernet frame plus additional protocol overheads. This would normally add up to 38 bytes (ethernet addresses, packet length, checksum, guard space, etc.)

38 bytes on top of a 1500 byte data packet is 2.5% extra. 38 bytes on top of a 40 byte ack packet is 95% extra, almost double.

So, for ethernet your measured ratio would be 1.95:49.025, assuming everything above; that's close enough to 2:50.

Now, there are two BIG caveats, and one little one.

Firstly, the cable modem is NOT ethernet. I really don't know enough about it to know what sort of overheads there are, but you can be sure there's a source and destination address. The fact that the cable side has MAC addresses which look just like ethernet addresses would suggest that they could be comparable.

Secondly, I've no idea at which protocol level the bandwidth limit is at. If it's discarding the lower-layer protocols before calculating the speed, then 1.33Mbps would be sufficient (although NOT enough to you to make best use of the bandwidth).

If the bandwidth limit INCLUDES the additional overheads (whatever they may be) then it's unlikely that 1.33Mbps would be enough: even 12 bytes (two mac addresses) is enough to increase the bandwidth requirement to 1.77M.

Thirdly, delayed-ack could reduce the ack traffic by half. It will for some people ... but from your calculations above, not you.

Of course, this is all academic if you want to do anything except downloading large files. Most people will be want do to things like check for e-mails, send e-mails, read web pages, have Windows doing updates, NTP updates, weather updates, RSS feeds.

More upload would be beneficial, but then again it's more than you'd get with 20Mbps so it's a move forward.

Would any reduction in Upstream bandwidth requirements be gained from use of things like Ack Suppression?

[foddy]

Originally Posted by broadbandbug

Would any reduction in Upstream bandwidth requirements be gained from use of things like Ack Suppression?

Yes, that's what I meant by the delayed-ack line. Of course, it could slow things down if it causes the sender to pass the end of its window.

Also, these are best-case figures (1500 bytes per packet). Not every path on the Internet has a 1500 MTU. Not every protocol is as efficient. You have to take into account request headers, for example.

If you assume that on average you need a 1:50 ratio and you're on the limit of upload speed, then for each 1k, you lose the ability to download 50k.

That doesn't seem like much, but HTTP headers can be that big - if a page is made up with dozens of small images, then it soon adds up.

[Chrysalis]

Originally Posted by TraxData

Yep, comcast are planning 20, 50 and 100mbit symetrical services for sometime late 2009.

I think the 100mbit will fall flat on its arse as their hardware isnt very good.

The rest are very viable though, the 50mbit symetrical is already being trialled (behind closed doors).

Just look at fios, 50/20 package...doesnt that make VM look pathetic?

I been wondering why the uk and usa markets are so different.

In the usa traffic shaping fell flat on its face with congress getting involved and there is clearly more investment going on in their broadband infrastructures (FTTH and DOCSIS3).

In the uk its quite the opposite we have providers falling over themselves trying to get the bottom customers (the ones who want the internet for nothing), there is no real investment going on other than in sales/pr departments and buying customers. Traffic shaping came without a bat from the media until it was already established, traffic shaping is now used by the majority of isps, sky are the only major isp not using it. In addition isps have got away with selling high speed products that in reality are not high speed. Because of this there is no motivation to upgrade the infrastructure to properly support high speed.

So why is this? Could be a few reasons here are some I can think off.

1 - Our consumers are stupid their main concern is price and as long as its cheap enough they fairly satisfied regardless of the impact on quality of service. Example talktalk starts free broadband suddenly people are ringing their current isp for a price match without understanding the economics of it.

2 - Lack of regulation/government/media attention, the ASA already gives isps the green light to call limited services unlimited, they consider a broadband connection that has a fibre backhaul but non fiber local loop to be described as a fibre connection. Both these things alone massively reduce the needs of an isp to upgrade its infrastructure as it now has a licence to missell its product, then we have ofcom who are only interested in a competitive market even if its means the consumer gets screwed. Example again, BT will not invest in FTTH or FTTC because ofcom want them to unbundle it cheaply which effectively makes it unviable to BT, if BT were rolling out FTTC/H then we can be sure VM would be reacting to it.

3 - Related to #2 somewhat our market is way too competitive, prices are too low, 8mbit broadband for £10 has no margins that can fund investment as well as making it no longer possible for light users to subsidise heavier users since there is no longer any profit on light users never mind heavy users, and VM is soon to be going down the same path when they upgrade all 3 of their tiers to faster speeds. It is basically mass selling of bargain basement broadband, quality goes hand in hand with price. Example, adsl prior to 8mbit adsl max was either 20:1 or 50:1 contention ratio and the 50:1 products were actually closer to 30:1, after adsl max launch we no longer see contention ratios defined in products and the reason for this is because now contention ratio is nearer 133:1 for a lot of adsl isps. There is no such thing as something for nothing, the price for increasing adsl speeds at no retail price cost was more people are now sharing the bandwidth. It will be the same on cable when the speeds go up.

4 - governments relationship with media copyright holders, we know they are backing them against the isps threatening to even legislate, so broadband packages with low upload limits make it harder to distribute copyright material.

Thanks to those who can be bothered to read it

---------- Post added at 02:13 ---------- Previous post was at 02:02 ----------

In regards to the turning off analogue.

Virgin was, however, unable to provide details of what will happen in individual areas where it currently provides analogue cable services but not digital. They include parts of Milton Keynes, Westminster, Southampton, Slough and Leicester.

I think VM may well withdraw from some areas especially now they ditched their LLU plans. It seems to make sense as to why they refusing to comment. Clearly they don't/can't upgrade these area to digital. I still remember the phone call I had with a VM exec PA who refused to say they will not withdraw from my area.

---------- Post added at 02:24 ---------- Previous post was at 02:13 ----------

Originally Posted by foddy

That DOES sound a bit simple ... at what layer were you measuring this bandwidth? From 1:46, I'd assume you're just measuring IP packets (40/1500 would be 1.15:46).

But, you probably performed this test over ethernet. On top of your IP packet, you have an ethernet frame plus additional protocol overheads. This would normally add up to 38 bytes (ethernet addresses, packet length, checksum, guard space, etc.)

38 bytes on top of a 1500 byte data packet is 2.5% extra. 38 bytes on top of a 40 byte ack packet is 95% extra, almost double.

So, for ethernet your measured ratio would be 1.95:49.025, assuming everything above; that's close enough to 2:50.

Now, there are two BIG caveats, and one little one.

Firstly, the cable modem is NOT ethernet. I really don't know enough about it to know what sort of overheads there are, but you can be sure there's a source and destination address. The fact that the cable side has MAC addresses which look just like ethernet addresses would suggest that they could be comparable.

Secondly, I've no idea at which protocol level the bandwidth limit is at. If it's discarding the lower-layer protocols before calculating the speed, then 1.33Mbps would be sufficient (although NOT enough to you to make best use of the bandwidth).

If the bandwidth limit INCLUDES the additional overheads (whatever they may be) then it's unlikely that 1.33Mbps would be enough: even 12 bytes (two mac addresses) is enough to increase the bandwidth requirement to 1.77M.

Thirdly, delayed-ack could reduce the ack traffic by half. It will for some people ... but from your calculations above, not you.

Of course, this is all academic if you want to do anything except downloading large files. Most people will be want do to things like check for e-mails, send e-mails, read web pages, have Windows doing updates, NTP updates, weather updates, RSS feeds.

More upload would be beneficial, but then again it's more than you'd get with 20Mbps so it's a move forward.

Yeah delayed ack is a big one, its on by default in windows XP and vista, I turned mine off for some reasons (mainly to decrease tcp latency) and I immediatly noticed my upload bw roughly double during downloading.

[spiderplant]

Originally Posted by foddy

That DOES sound a bit simple ... at what layer were you measuring this bandwidth? From 1:46, I'd assume you're just measuring IP packets (40/1500 would be 1.15:46).

I was measuring at the ethernet frame level (dest MAC up to, but possibly not including, FCS). But the all-important bit is that it was indeed delaying ACKs. I didn't take too much notice, but I think the data to ACK packet ratio was around 1.6.

[foddy]

Originally Posted by spiderplant

I was measuring at the ethernet frame level (dest MAC up to, but possibly not including, FCS). But the all-important bit is that it was indeed delaying ACKs. I didn't take too much notice, but I think the data to ACK packet ratio was around 1.6.

OK, that sounds fair. So your ACK packets would have been 54 bytes? 78/54 is 1.44. 1/46*50*1.44 = 1.57.

For *ethernet*, you're still over 1.5M. Does anyone have in-depth knowledge of the framing used on cable modems, and how exactly the bandwidth is limited?

[spiderplant]

Aren't you ignoring the delayed/suppressed ACKs? I agree it would be over 1.5 meg if every data packet was acked. But diviide by 1.6ish (or 2 on a good day) and you're well under.

AIUI DOCSIS adds an extra 6 bytes to the ethernet frame, but I don't know at what level the rate limiting works.

[foddy]

Originally Posted by spiderplant

Aren't you ignoring the delayed/suppressed ACKs? I agree it would be over 1.5 meg if every data packet was acked. But diviide by 1.6ish (or 2 on a good day) and you're well under.

AIUI DOCSIS adds an extra 6 bytes to the ethernet frame, but I don't know at what level the rate limiting works.

Nope, I was taking your measured figures (1:46 taking delayed-acks into account) and multiplying by 1.44, which is the additional overhead on ethernet for ack packets (78 bytes total / 54 measured).

6 bytes on top of the ethernet frame is much less than the 24 I added, but then it might include the CRC as part of the frame, which would make it 10.

It's all theoretical, of course.

[spiderplant]

Fair enough. Let's wait and see.

[|Kippa|]

Quite a few of you say that 1.5mbit would not be enough to download at full 50mbit. What upload speed would be needed to achieve the minimum 50mbit download rate? Would it still be sub 2mbit?

[popper]

http://www.cableforum.co.uk/board/12...l#post34545407