MWEB 40Mbps bonded ADSL held back by Telkom

[fragtion]

Not its not. Why do assume that the customer with 4 (bonded) lines will use more than the 4 individual ones?

I didn't say that a customer with 4 bonded lines would use MORE than if those 4 lines were installed on individual premises, I just said that BY them using (read: maxing out) all 4 those lines simultaneously, they effectively cancel out a large portion of contention ratio provisioning. Unfortunately I'm not too clued up about the maths or I'd work it out for you, but I'm most certain about the principle. Just look at my previous post... let's assume a Telkom exchange has a 50Mbps pipe to ESR, and 50 adsl ports... Now if one customer has a 40Mbps bonding solution and initiates one download which utilizes all 4 those lines, the whole remainder of ports will be left with 10Mbps... That's all I'm pointing out here: I stated that I'm aware that many exchanges (for example the Pierre van Ryneveld exchange) are over-contended as it is, so allowing many customers per exchange to bond is just totally going to kill it even more - think industrial areas where bonded solutions are most in demand... I have no idea how those exchanges could cope during business hours? Suddenly the title "Mweb 40Mbps bonded solution held back by Telkom" seems to make sense, as much as it's against my will to defend Telkom...

 

[Roman4604]

Now if one customer has a 40Mbps bonding solution and initiates one download which utilizes all 4 those lines, the whole remainder of ports will be left with 10Mbps

Yes but statistically its swings & roundabouts.

Ignoring the fact that each DSLAM port has the same priority at any one instant of time, the 40Mbps customer will complete the download in a quarter of the time, creating unused time for the others to max their lines. Commecially its also the same, the one with 4 lines is paying 4 x the amount as one with a single line.

It doesnt change the equation at all.

 

[MickeyD]

I didn't say that a customer with 4 bonded lines would use MORE than those 4 lines, I just said that BY them using (read: maxing out) all 4 those lines simultaneously, they effectively cancel out a large portion of contention ratio provisioning. Unfortunately I'm not too clued up about the maths or I'd work it out for you, but I'm most certain about the principle. Just look at my previous post... let's assume a Telkom exchange has a 50Mbps pipe to ESR, and 50 adsl ports... Now if one customer has a 40Mbps bonding solution and initiates one download which utilizes all 4 those lines, the whole remainder of ports will be left with 10Mbps... That's all I'm pointing out here: I stated that I'm aware that many exchanges (for example the Pierre van Ryneveld exchange) are over-contended as it is, so allowing many customers per exchange to bond is just totally going to kill it even more - think industrial areas where bonded solutions are most in demand... I have no idea how those exchanges could cope during business hours? Suddenly the title "Mweb 40Mbps bonded solution held back by Telkom" seems to make sense, as much as it's against my will to defend Telkom...

Your business areas have better infrastructure than your residential areas.

 

[fragtion]

Commecially its also the same, the one with 4 lines is paying 4 x the amount as one with a single line.

It doesnt change the equation at all.

From Mweb's point of view, of course not. From Telkom's point of view, they MUST ensure that their exchanges can support this simultaneous increase in load. Imagine if we all decided to fill our swimming pools tomorrow between 2pm and 4pm - clearly the primary supply pipes would not cope and as a result everyone's hose-pipe would trickle water at a fraction of the normal rate. The same concept applies to a bonded solution where multiple lines are utilized simultaneously, disrupting normally expected contention ratios. Yes it is true that with bonding a given download would finish quicker, but the difference is that a good percentage of bonding clients intend to utilize the majority of bandwidth for prolonged periods... on the contrary you're trying to make me believe that it's all short bursts - if that were the case then there'd be no need to bond, if it's a super quick download on 16Mbps then it won't be the end of the world of a wait on 4Mbps either... so why bond then?

 

[Roman4604]

but the difference is that a good percentage of bonding clients intend to utilize the majority of bandwidth for prolonged periods

On what basis would you assume bonded customers intend to use their lines for more prolonged periods than the average for single line customers?

so why bond then?

Because you desire/require, and are willing to pay for, more speed that what a singular line is able to offer.

 

[fragtion]

On what basis would you assume bonded customers intend to use their lines for more prolonged periods than the average for single line customers?

The bonding customer's requirements are far above normal anticipation for an average 4Mbps customer which until recent exchange upgrades was Telkom's highest available ADSL offering. They wouldn't upgrade 4Mbps customers to 10Mbps until upgrading the exchange's overall backhaul capacity (the upgrade to metro-ethernet) - ask: why is this? It's because Telkom would not have provisioned for such a steep simultaneous spike in traffic from just one customer at a given time, as this disrupts contentions for their overall bandwidth capacity. Why don't Telkom offer an 'up-to 24mbps' service, or 100Mbps fibre-to-the-home? The answer to that question would be the same justification for why they would frown upon bonding: the last-mile local-loop infrastructure can't accommodate such large independent spikes in demand from a single customer at this stage, so circumventing the problem with solutions such as bonding will in theory cause a degradation in service for other customers... the truth is they just don't have the capacity to meet the demands of customers bonding 4Mbps lines as effortlessly as they'd meet the demands for an equal amount of independent 4Mbps lines, because the independent lines would not saturate the transit simultaneously in the way that the bonded lines would. Of course if the overhead bandwidth amount was more (contention ratio was less) then it would average out so that the difference between bonding and individual lines wouldn't really make a difference as you point out... but we all know that Telkom's exchanges don't exactly have this kind of bandwidth supply from their exchanges to the PoP/edge routers =/

 

[Roman4604]

because the independent lines would not saturate the transit simultaneously in the way that the bonded lines would

Again your assuming.

For example, on average why would an office of 40 people ustilising a 16Mbps bonded services cause more load over the course of the month than 4 offices of 10 staff, each utilising a 4Mbps service?

Either way the average contention experienced on the DSLAM's backhaul is the ratio between the access (n x 4Mbps access ports) and the backhaul capacity. Bonding does not change this ratio, but upgrading the individual port speeds to 10Mbps does (dramatically).

 

[fragtion]

Again your assuming.

For example, on average why would an office of 40 people ustilising a 16Mbps bonded services cause more load over the course of the month than 4 offices of 10 staff, each utilising a 4Mbps service?

Either way the average contention experienced on the DSLAM's backhaul is the ratio between the access (n x 4Mbps access ports) and the backhaul capacity. Bonding does not change this ratio, but upgrading the individual port speeds to 10Mbps does (dramatically).

For ordinary usage, I'd tend to agree with you, but the problem is there is a lot more bandwidth available on demand to a customer with 16Mbps bonding (not to mention the 40Mbps bonding proposed by this article which Mweb wishes to provide) than there is to each of those 4Mbps offices, and that's where the problem comes in: saturating a 40mbps line is a very different story for an exchange when compared to saturating an equal amount of 4Mbps lines but at different independent time-frames so that the sumtotal of bandwidth used by the individual lines on average at any given time is less than the sumtotal of a peaked bonded solution. Unless there is sufficient bandwidth overhead, this could be problematic.

Consider:
4Mbps Line at office A makes a full 4Mbps download from 10am to 11am
4Mbps Line at office B makes a full 4Mbps download from 11am to 12pm
From 10am to 12pm, the exchange not once exceeded 4Mbps despite two customers utilizing a full 4Mbps at different times. That's the basic idea behind contention ratios... and even though for practical cases the usages do overlap, you can still work out an average expected peak and then roll-out services based on that. It's exactly this expected peak which is disrupted by bonding, or as Profmerlin said: also by uncapped. With uncapped guys tend to download for much longer sustained periods, thereby pushing the expected contention ratio boundaries, because more lines are maxing out simultaneously (overlapping) rather than most being idle (larger free bandwidth pool); similarly, with bonding, these boundaries are pushed beyond what Telkom would have anticipated and provisioned for, and therefore at the end of the day customers experience an overall degraded service and the exchange must be upgraded to cope with the higher demand surges which may arise from just a few customers who now have this supercharged bandwidth at their disposal.

Let's say back in 2003 a Telkom exchange had 10Mbps overall capacity to serve a couple of 512k clients. This would not be a problem because the probability that all of those users will collectively reach that maximum download of 10Mbps simultaneously, is low (assuming there aren't too many, of course). If some of these clients were to bond these 512k lines to obtain say 4-10Mbps (but still having the same overall number of 512k lines to that exchange), then they could very quickly reach the capacity of that exchange by themselves (breaking Telkom's contention ratio expectations), so then the odds of reaching that 10mbps is increased because that improbable potential outcome is essentially made more probable by enabling a smaller group of users to essentially do so by themselves on demand. A greater backhaul capacity then needs to be deployed to counteract the higher probability->frequency of that bottleneck being reached. Now just scale that up a notch for 4/10Mbps where the exchange's maximum capacity is in the same basic ratio as the given example - the same applies...?
I've even heard unconfirmed reports that some Telkom exchanges offering up to 4Mbps DSL still today have as low as 10Mbps total throughput to the ESR. This is an extreme situation, but here a SINGLE user trying to get 16Mbps with 3 bonded lines would not even get 16Mbps, yet that exchange would cope fairly easily with even more than 3 independent 4Mbps customers, without them even knowing that their exchange is so limited/shared, because you'd need at least 3 x 4Mbps lines maxed out simultaneously in order to surpass that capacity - a much less likely event if you lower each user's maximum throughput (and then you can in turn provide more users with a service).

So my 'assumption' does seem justified in that it's based on interpolating potentially likely outcomes in reality ;D

 

[Roman4604]

So my 'assumption' does seem justified in that it's based on interpolating potentially likely outcomes in reality ;D

Only because you paint the best case scenario for the individual line users i.e. each waits for the other to finish their download. Just as improbable as every access request from the bonded customer must be maxing out all 4 lines.

Anyway I dont see what what your motivating for? That bonded is bad and should not be allowed as it adversely affects single line users?

If so one could could make a similar argument that 4Mbps users have an even more pronounced adverse affect on 384Kbps users, therefore all connections should be reduced to the lowest common denominator ensuring zero probablity of any unfairness.

Ultimately if the collective access demand is oversubscribing the avaliable backhaul, the network operator must upgrade the backhaul. After all they are being paid handsomely to do so (twice over).

 

[fragtion]

Only because you paint the best case scenario for the individual line users i.e. each waits for the other to finish their download. Just as improbable as every access request from the bonded customer must be maxing out all 4 lines.

As you said, they get bonding because they want more speed, so they can download a large file in a shorter period of time. That response shows that it's not improbable that a bonded customer WILL max out all 4 lines at once simultaneously on demand. It therefore reinforces my point that it's more probable that those 4 lines would be maxed out simultaneously with bonding, than if they were assigned to 4 individual customers who would IN EFFECT make use of the lines at different times.

As for 'best case scenario for the individual lines, ie each user waits for the other to finish their download' - I'm fully aware of that, and I specifically pointed out the following:

That's the basic idea behind contention ratios... and even though for practical cases the usages do overlap, you can still work out an average expected peak and then roll-out services based on that.

The only reason I had to resort to an extreme 'ideal' (and yes, improbable) situation where the individual usages NEVER overlap, was to make the point clearer to see (The same reason why I used 2003 512k ADSL as an example): it simplifies it for understanding, something I was seemingly forced to do based on your persistence to negate a valid point on my part

Anyway I dont see what what your motivating for? That bonded is bad and should not be allowed as it adversely affects single line users?

In reality the probability that all individual users will be reaching a collective maximum throughput by simultaneous downloading is low, as opposed to more easily reaching that limit by bonding lines so that one or a few users can max out multiple lines on demand. That was my initial contention which you didn't accept (see page 3) - it was initially a simple statement which later had to be broken down and justified. To answer your question more directly: within the context of the current state of Telkom's infrastructure, yes bonding is bad and should not be allowed as it adversely affects single line users, especially on exchanges where throughput to the ESR is not in an ideal situation to cope as the bonding disrupts the mean expected usages as would be clearly visible on a graph for that exchange's uplink bandwidth usage

If so one could could make a similar argument that 4Mbps users have an even more pronounced adverse affect on 384Kbps users, therefore all connections should be reduced to the lowest common denominator ensuring zero probablity of any unfairness.

Of course you could, but this just reinforces what I said: Telkom specifically didn't roll out 10Mbps until exchanges were upgraded; they can't be too happy then about customers bonding to reach 16Mbp because that is beyond their 'maximum predicted per-client usage'. So we take it a step further: 40mbps bonded? "MWEB 40Mbps bonded ADSL held back by Telkom" then seems like a very logical decision by Telkom, unless we want absolute crippling of the exchanges. One can't help but get the impression that Mweb seems to think that bandwidth is in abundance on Telkom's last-mile local-loop... well as much as we'd all like that to be true and support Mweb about it, we all know that this is far from the reality...

Ultimately if the collective access demand is oversubscribing the avaliable backhaul, the network operator must upgrade the backhaul. After all they are being paid handsomely to do so (twice over).

Of course, but don't forget that's another best-case painted scenario right there to assume that every Telkom DSLAM port must be guaranteed its bandwidth equivalent for that port's sync profile (after all, this is a key reason as to why Telkom explicitly advertises it as a 'best-effort' or 'UP-TO 10mbps' service, apart from the distance to exchange factor)... nowhere in the world is that the norm, in light of contention ratios - in fact the standard contention ratio is 20:1 which means 20 x 4Mbit customers would share a single 4Mbps pipe just fine under normal conditions, It's not difficult to see from this how bonding has a negative effect especially if multiple bonded clients utilize their already bonded lines in unison, so that the potential of having reached that total throughput would be much less if each of those unique 4Mbps lines were designated to individual customers instead

 

[F4ckn]

In theory the back hauls should have enough capacity for the amount of lines that telkom dish out. (Even though this may not be the case in all situations, it sure is in most of the metro linked DSLAM's which telkom are upgrading all the time)

I think bonded has its place as at the moment we don't really have any other high bandwidth last mile deliveries in place and there is a need for it.

This argument can go on for ever but the point is your arguing about dynamic circumstances dependent on the area and the need.

Just my 2c chaps

 

[fragtion]

A passive median stance/position is always the safest ;D Anyways we both made our points, Roman is correct that one or a few bonded solutions should not pose a problem and that even with bonded solutions, the same principles of averaging do apply (swings and roundabouts); but I'm confident that I'm also correct when saying that bonding does manipulate contention ratios and that if the exchange backhaul is not sufficient, then reaching that limit becomes way more possible with bonding, especially if 40Mbps bonding becomes a reality... EEK ;D

 

[Roman4604]

Anyways we both made our points

Ya well, we'll just have to agree to disagree, however I do worry about the high level premise of your argument.

If I understand you correctly; service providers should not be allowed to bring new offerings to the market that push the envelope of performance/capability, as it may expose deficiencies in exchange backhaul capacities. In essence innovation must be retarted down to Telkom's pace of development.

Yes, exchange congestion is a ballache for all, but in my view this a dark & slippery path to advocate. With this mentality we'd all still be hitting our 3GB caps somewhere near the begining of the month.

If we are to get anywhere near our overseas brethren, industry must allowed to innovate, if fact encouraged to accelerate at it. Telkom must just get their act together and keep pace (as mentioned they still take a significant share of the income with LR+IPC charges).