DSL Speed question?

[1966]

OK, I know I think to much sometimes but got to thinking HOW DOES AT&T control the speed of the various dsl services? Their Fast Access dsl lite is at one price, then dsl ultra and supposed to be faster is another price. Do they do it with software or hardware or is there really a difference before I jump off and do this?

[mjc]

Well, one method of control is distance from the equipment. The farther you are located away form the CO the lower the maximum speed you can achieve will be. Other than that, it is a combination of hardware and software. Kind of like the old dial up days...28.8 vs 56 k modems.

[tommy]

HOW DOES AT&T control the speed of the various dsl services?

By inserting simple electronic filters on the phone line at the appropriate places. The filters control the bandwidth that can get past the filters for the DSL portion of the line. And then they do just the opposite (with filters) with the phone portion of the line so that you don't hear the DSL signal on the phone (when both are in use at the same time.)

[Variable]

No, the bandwidth is not filtered by a filter on the phone line. It is rate limited at the router. It is done through software not hardware. The commands depend on the router. But it is usually applied to a interface or virtual interface. The most common rate limit is drop. So that when the rate limit is exceeded the packets are dropped. The filter that you would put on the phone line, or more likely now put outside the home is to separate voice from data.

[Paul Komski]

I think the poster poses a very good question. I have an incomplete understanding of things but here goes.

All other layers sit on layer-1 of the OSI or TCP/IP model. Layer-1 is the physical layer that involves the actual transmission of a data bitstream; a layer that is thus deeply associated with physical hardware and transmission media. This data bitstream is in a sense raw and as of yet unmodified into packets or otherwise "manipulated" by software.

The transmission rate of any datastream will always be limited by the "weakest link in the chain". In the physical layer this must be in the "ports/network nodes" at either end and as an intrinsic feature of the medium or bus itself; (a gigabit ethernet would not be as fast as a 10 gigabit ethernet). In serial communications such a bitstream doesn't need to be split at one end and recombined at the other - and is just one of the reasons for the recent migration to serial communications as "the norm".

Routers are at the heart of all internet communications and, in a simplistic way, are responsible for receiving and re-sending data, made intelligible by software reading the next layer up where the raw bitstream has been "organised" into packets (which include an element that must define a destination).

The data then gets transmitted just as fast as the medium and ports and buffers will allow. If more than one destination is actively involved (the contention ratio I believe its called) then the slower will the communications be to any one particular destination. A router with lots of potential addresses is thus liable to transmit slower to any one destination than one that can only transmit to a few addresses.

Of course that router can only transmit as fast as the information it receives (and which has come from another router). The aorta, if you like, can only pump out as much blood as the venous supply that it gets.

I guess that there is one of two likely ways for an ISP to determine/control (all other things being equal) what bandwidth they provide to their clients. One is to change the venous supply by using different media, different routers and so on or to limit the arterial delivery by applying some kind of filter on the outlet side. A filter that would be likely to be in software that controlled how many packets could be output in any unit time.

I am not a network guru by any stretch of the imagination but perhaps my own take on things can be the basis for any real understanding or discussion of exactly what goes on and allow an understaning of how the ISPs can thus be able to charge differentially for the different bandwidth services that they provide; provide down the same medium be it ADSL or wireless or cable or whatever.

Variable may already have said everything necessary eloquently and in just a few words; that bandwidth is merely controlled in software by the ISP's routers dropping packets to limit the bandwidth. Or is there more to it?

[1966]

This thread has turned out to be pretty interesting. I have installed Bellsouth's
dsl service( now AT&T ) on one computer so far for a friend when we had to re install and reformat her computer. The service comes with a modem that as I understand it does work as a router also and software. The software actually did the installation and did it quite well in my viewpoint. The software looked at the network card and had several questions each with a choice of answers. There was a suggested answer based on the network card installed which happen to be an older Aopen brand card. The software also found a problem with the card and corrected it. After answering some basic account questions and putting in the suggested ip address I believe it was, the machine rebooted and bingo the internet was running on dsl and very quickly I might add. My point with all of this is I have always assumed that software controls the speed as well as everything else but my ole mind just got to wondering "does it".

[mjc]

Other than the initial limits on the distances involved, not really. DSL is one of those things were distance and line quality really matter.

[Variable]

that bandwidth is merely controlled in software by the ISP's routers dropping packets to limit the bandwidth. Or is there more to it?

That's it. That is how it is done. I can create a group; assign that group 2mbps up/down or 2mbps up/4mbps down. I can then add your PPPoE authenticated username to that group. When your modem connects, it is authenticated, rules are applied based on whatever criterion I apply to your group, user and port and that is it. If you had a dedicated connection, i.e. you had a cable directly connected to my core routers; I would apply the restriction on your interface i.e. the port your cable was plugged in to. With broadband, your client’s requests will be aggregated over one (or more) physical cables. So I can then apply rules based on your initial authentication, etc. or even by your IP. But with any consumer grade broadband your authenticating to login to the core routers, the routers that provide you bandwidth, during that authetication (RADIUS in many cases) that is where I will stipulate what group you are assigned to. That group can have preset bandwidth rates and any other rules I care to apply (like additional rate limiting rules or blocking for file sharing ports) This is common, it eases back end administration. You structure rules based on the tiered business offerings.

You would not limit the link speed i.e. you would not auto negotiate 10mbps full or half duplex because it is not really useful. 10mbps is still MUCH faster than any DSL could achieve. If you were a client in the datacenter I work at, and I wanted to rate limit you to 10mbps. I would still do it on the router, when your server connects it will show a link speed of gigabit (if you have a gigabit NIC of course) but you could not sustain that speed if I rate limit your port at 2mbps.

[Paul Komski]

Good stuff. Thanks.

DSL is one of those things were distance and line quality really matter.
Just wondering. Does the bandwidth fall off with distance, per se, or because those clients nearer the source get preferecne; or both maybe?

[Ghost_Hacker]

All other layers sit on layer-1 of the OSI or TCP/IP model. Layer-1 is the physical layer that involves the actual transmission of a data bitstream; a layer that is thus deeply associated with physical hardware and transmission media. This data bitstream is in a sense raw and as of yet unmodified into packets or otherwise "manipulated" by software.

That is not the correct picture of how it works. The way to look at it is that data flows down the OSI model at the source and up the OSI model at the designation. So by the time data reaches layer 1 at the source it has already been manipulated into packets at the network layer which will become frames at the datalink layer and bits at the physical layer. Think of it as sorta like the Russian dolls that fit a smaller one into a larger one. Once data reaches the designation it goes back up the OSI layers so bits become frames which then become packets. ( Please note that the OSI model and the TCP/IP model are NOT the same though you can map the TCP/IP model to the OSI one)


The transmission rate of any datastream will always be limited by the "weakest link in the chain". In the physical layer this must be in the "ports/network nodes" at either end and as an intrinsic feature of the medium or bus itself; (a gigabit ethernet would not be as fast as a 10 gigabit ethernet). In serial communications such a bitstream doesn't need to be split at one end and recombined at the other - and is just one of the reasons for the recent migration to serial communications as "the norm".

At the physical layer ports do not exist. Ports exist at the tranport layer. Bitstreams are not recombined, they can be retransmitted if lost, but they are not "spit apart". In ethernet networks physical frames have a start and an end they can be short or long but they are never "split" though they may be padded to meet a minimum size. Don't confuse frames moving thru a cable with data moving thru a network. Data packets ( or datagrams) can be split and recombined at a designation host this is normal for TCP/IP networks and is part of the protocol standard.

Routers are at the heart of all internet communications and, in a simplistic way, are responsible for receiving and re-sending data, made intelligible by software reading the next layer up where the raw bitstream has been "organised" into packets (which include an element that must define a destination).

Again the packets are already there just hidden in the frames. A router will open a packet to read the designation host IP address, rebuild the packet and drop it into the output q of the correct interface.


The data then gets transmitted just as fast as the medium and ports and buffers will allow. If more than one destination is actively involved (the contention ratio I believe its called) then the slower will the communications be to any one particular destination. A router with lots of potential addresses is thus liable to transmit slower to any one destination than one that can only transmit to a few addresses.

Routers will drop packets as their buffers get full and they can't process any futher packets. This cause retransmissions to occur which can be seen by the end user as a "slow down". If configured correctly routers should never have "lots" of potential address. Remeber routers route between NETWORKS not host so they don't have address per say.

Hope this helps :)

[Paul Komski]

Thanks Ghost it both helps and hinders. Before moving on to the "Russian Dolls" and any layering models I wish I could first of all visualize/comprehend what is happening at the physical level - and how the concept of frames is applied. I thought I had an understanding of how "data", for want of a better word, was transported along various media but once I saw this question posed I realized that I didn't really understand after all.

With radio (if my thinking is right) there is a RF signal traveling at the speed of light at specific "wavebands". The underlying sine wave is modulated by frequency or amplitude so that a signal transmitted at one end can be received at the other end. An analogue signal can be converted in real time via an analagous receiver/transceiver. A digital signal would need to be coded and decoded having used FM or AM waves as the carrier.

If that is a reaonable understanding then an equivalent signal can also be transmitted in copper/metal wires using very high frequency AC of low current and thus also carried at great speed. Presumably pulses of light are modulated in similar fashion in optical fibre transmission, where the huge advantage is the very large number of "lines" that can be carried in one cable.

The "physical port" at the "listening end" would commonly be a transceiver chip of one sort or another whatever medium was used. The transceiver allowing the digital signal to be interpreted by software. Only at this point do the other layers, frames, packets and so forth come into play.

If it is taken as read that such physical signals are transmitted at or nearly at the speed of light then there are no strictures, per se, placed on the speed of transmission. The limitations start to come into play when data is scrambled or is incomplete and a request needs to be made to resend the data (TCP) as opposed, for example, to UDP transmission which is only one-way.

Is this a reasonable understanding of the physical layer?

[Ghost_Hacker]

...The "physical port" at the "listening end" would commonly be a transceiver chip of one sort or another whatever medium was used. The transceiver allowing the digital signal to be interpreted by software. Only at this point do the other layers, frames, packets and so forth come into play.

If it is taken as read that such physical signals are transmitted at or nearly at the speed of light then there are no strictures, per se, placed on the speed of transmission. The limitations start to come into play when data is scrambled or is incomplete and a request needs to be made to resend the data (TCP) as opposed, for example, to UDP transmission which is only one-way.

Is this a reasonable understanding of the physical layer?

You almost have it, but there are several other factors at play.

Yes, the signal moves at the speed of light, but frames (or cells or whatever physical transport package the media uses) only move at the frequence "speed" agreed upon by the hosts transmitting the data.

Every physical media type has specs which effect the frequence speed and distance it works best at. These specs effect the strength of the transmited signal. Lets look at Fastether net (100Mps) as an example.

Fastether net uses Cat5e cable (Most cat5 cable now days is in fact Cat5e). Cat5e cable can support a frequence range of 1- 100MHz, so data can move no faster then 100 Mhz on this cable and still be "seen" as data and not as junk by the hosts using this media.

Another factor that effects the speed of the frames moving across a Fastether net cable is crosstalk (NEXT as it is called happens at the point where the cable connects to the "jack" allowing the signal from one cable to bleed into the other cables) NEXT is an error condition which causes signal loss. Higher rated cables have better "NEXT" performance then lower rated cables and can therefore support faster frequences and move data faster across your network.

Every Media type can suffer from "line noise" which can futher degrade signal strength and slow down a network. In the ethernet world this is measured using what is called an SNR margin(signal-to-noise ratio). A higher SNR is a good thing when comapring cables. (though unless your talking Gigabit cable you don't need to worry about it much) cause it helps to combat problems from ambient noise (noise caused by power lines for instance).

The last factor for Fastether net cables is Transmit Echo Noise. TEN, as I will call it, happens in full duplex networks. TEN is when the outgoing signal on a set of wires that make up a cable hops over into the incoming signal on the incoming wires of that same cable. TEN can cause loss of signal and therefore effect the speed of the circuit.

These are some of the phyiscal problems which might effect the speed of any given fastether network circuit, but all circuits have their "signal probelms". Fiber, for instance has it, "modes" different moded fiber can support faster transport speeds.

Does this help explain it better?

EDIT also remember that the problems you mentioned (incomplete packets, srambled data) do not occur at the physical layer. The physical layer knows nothing about "data". It knows only about the signal ( the stream of bits across its media) theses bits can be made unreadable for whatever reason and the physical layer can see and correct for this. But only so long as the errors are within norms for the media in question.

[Paul Komski]

Thanks Ghost - it all helps - and is all food for thought. I'm never content to just know that a thing works (or doesn't of course) and get almost obsessional about really understanding things from the ground up. I need to think much more about this and as ever there is a lot of ancillary "homework" to do. I have always kept networking on the long finger (to use an Irish colloquialism) but am being forced more and more to really get to grips with it.

[Ghost_Hacker]

You have a good understanding already. You just need to fill in some details here and there :)

To help you I would think about using the Russian nesting dolls as one way to picture how data moves thru the OSI model. It really does help to simplfy and picture how the data from one layer is encapsulated into or becomes the data for the next layer when data is going down the OSI model and the reverse when data is passed up the OSI model.

Example:

At the source a packet from the network layer is placed into a frame with a datalink/LLC header at one end and a FCS "footer" at the other with the data from the network layer between them.

That frame is then encoded by the phyiscal layer (Manchester encoding for Ethernet) with a preamble placed at one end of the frame and a CRC at the other. Please note that the preamble is not read into the NIC's input buffer but is used to sync up the timing or clock between the two interfaces. The CRC of course is used for error checking. So in this case you can see that a smaller "doll" is placed into a larger one as we go down the OSI model.

At the designation both the preamble and the CRC are "thrown away" and the frame is extracted and passed up to layer 2 where the datalink/LLC header and FCS foot are thrown away,the IP packet extracted and sent up to layer 3. Again in this case we see how a larger doll is opened to get the smaller doll inside as we go up the OSI model.

Hope this helps some :)

EDIT Also I would note that you should never mix and match error types or "terms" with the wrong layer. So when talking about the physical layer don't mention ports as ports exist at layer 4 not 1. Things like that are not really important to a home gamer and sometimes it can't be helped when explaining things, but they do help to knock your networking game up a notch. :D

[Paul Komski]

So when talking about the physical layer don't mention ports as ports exist at layer 4 not 1.Yes - it was loose talk. The whole concept of ports means different things in different contexts and this is of course in a networking context.

I suppose I was using the term as being any junction between one medium and another with the analogy of changing from land to sea travel at a sea port. On electrical/electronic equipent any socket and plug is equivalent to this type of physical "port". I used once to be very confused by the term port as defining, by numbers, which internet protocol used which port until the penny dropped that these are virtual and not physical - at least in any meaningful sense to me.

[jlreich]

Paul, have you read through a good description of the OSI model? Have a look here (http://www.cisco.com/univercd/cc/td/doc/cisintwk/ito_doc/introint.htm#wp1020580), it will help give you a good idea of what each layer does and how data is moved down and up through the different layers.

[Ghost_Hacker]

Yes - it was loose talk. The whole concept of ports means different things in different contexts and this is of course in a networking context.

I suppose I was using the term as being any junction between one medium and another with the analogy of changing from land to sea travel at a sea port. On electrical/electronic equipent any socket and plug is equivalent to this type of physical "port". I used once to be very confused by the term port as defining, by numbers, which internet protocol used which port until the penny dropped that these are virtual and not physical - at least in any meaningful sense to me.


Yeah I thought you where just using the term to mean something else other then its "networking" def.

I had the same issue with ports and alot of protocol terms till I started to study the actual make up of a TCP/IP packet and compare that to say a site like jlreich linked to. It helped cause in studing the packet layout you could actual see where a layer of the OSI wrote/read its data. It became less abstract.

Anyways, like I said you already have a nice understanding better even then some IT professionals who can tell you the layers but would not know what each really does to save their skins LOL :D

[Paul Komski]

Thanks jlreich - I have dipped into Cisco before for other sorts of information - they are pretty authorative to say the least.

Being able to interpret the packets is of course a key to understanding much of what is going on - particularly in software. In a simplistic way that is just to do with data i/o to/from the packets.

If one just takes it as read that the packets are checked, buffered, re- requested as necessary, re-routed, re-assembled and so forth then interpreting the transmitted data is simple enough in the "upper layers" of the OSI model. That is in itself interesting enough but it is the physical layer and (now that it has been better defined) the data link layer that are currently of the greatest interest to me.

Its what the electrons and any EMR are doing that is one fascination. What happens to the binary data also needs understanding but it doesn't involve a comprehension of applied physics.

Anyways all the inputs have been helpful and maybe after Christmas I will have more time on my hands to do some proper reading.