ADSL capacity

[rookie1010]

Hello
I was wondering what the advatage of ADSL to operators is?

what i mean to ask is since it is the same line running between subscriber and provider, what is the advantage of providing 128Up and 256K down, say?

i mean why not provide 256 bothways?

 

[JackMDS]

At the end, when you tally the total Traffic.

256 + 256 = 512 x 100,000 (users) = 5,120,000 generated traffic :thumbsdown:

256 + 128 = 384 x 100,000 (users) = 3,840,000 generated traffic :thumbsup:

Traffic (bandwidth) cost money the Internet system is not running free by some benevolent aliens.:shocked:

95% of Home users use the upload for Browser clicks and email and they can do even with slower upload bandwidth.

If you really need to run a server or exchange huge files for professional or business purposes, it means that you are a Real Power User.

Well, Powerful people can effort to pay a little more and upgrade to the next level of DSL Service and you would get what you need.:thumbsup:

:sun:

 

[spidey07]

Its to prevent running servers.

That's really all there is to it.

That and the equipment providing the service is much higher powered and can "drive the line" towards you where as to get the same kind of push on the upstream would have a significant cost to the consumer (not a good thing)

 

[randal]

Also, depending upon the flavor of DSL, frequency allocation on the copper pairs limit the available amount of bandwidth available for upstream & downstream communications. Combine that with the distance limitations of ADSL and every ISP's desire to focus on stability over speed, and you end up with asymmetric service. Right now, most ADSL implementations max out at 896kbps upload, and with the use of some neato technologies is hitting 7mbps down, although 3-5mbps down is much more common.

The upstream portion of the frequency space is very close to the voice range on a line. A LOT of providers configure things so that it uses even less of the upstream frequency band so as to avoid any possibility of interference with voice calls, at the expense of upstream bandwidth. As I mentioned earlier, many ISPs will sacrifice bandwidth & latency to ensure a stable link, utilizing methods like interleaving, forward error correction, etc -- you may not get the most bandwidth ever, but you're also not spending 9385 hours on the phone with tech support, which is their primary concern.

Also, differentiated services make ISPs a lot of money. As the previous posters said, bandwidth isn't free ... even though all the upstream links the ISP has are symmetric, the uploading is "more expensive" simply because they want it to be. Do keep in mind that the difference between your upload & download speeds is often sold to colocation, dedicated servers and other customers that regularly -serve- content instead of retrieve content -- ensuring that the ISP gets the most out of every kb of bandwidth they have available.

Right in the middle of a medium-sized ADSL deployment,
randal

 

[rookie1010]

so the two reasons for assymetric operaton is

istance limitations of ADSL on the Uplink (link budget?)

"The upstream portion of the frequency space is very close to the voice range on a line." Does this mean that there is more bandwidth devoted to downstream than upstream. i guess i had just considered the last mile aspect and thought that since the cable is yours why not use it to the max.

what are colocated, dedicated servers?

 

[randal]

Originally posted by: rookie1010
Distance limitations of ADSL on the Uplink (link budget?)

"The upstream portion of the frequency space is very close to the voice range on a line." Does this mean that there is more bandwidth devoted to downstream than upstream. i guess i had just considered the last mile aspect and thought that since the cable is yours why not use it to the max.

what are colocated, dedicated servers?

If you're a wireless guy, then the term "link budget" is closely related to the asynch nature of ADSL.

Yes, the frequency space allocated to upstream and downstream differ quite a bit. Using DMT+FDM, upstream gets from 26kHz to 138kHz (112kHz usable) while downstream gets 138kHz to 1.1MHz (962kHz usable); that discrepancy is the technical limitation to getting symmetric service out of ADSL.

They cannot "use it to the max" because of other issues as well -- any lower frequency would interfere with voice, and any higher of a frequency would result in serious attenuation (power & signal loss); those two little unshielded copper wires only work so well. The frequencies used were agreed on by national (ANSI) and international (ITU) standards bodies so as to avoid interference, foster compability and to provide the best solution engineering-wise.

Colocated & dedicated servers are basically machines that are physically located inside an ISPs datacenter and directly hooked to a network, compared to being at the end of a comparitively slow DSL/T1/DS3. 100mbps of LAN connectivity inside a datacenter is generally better and way cheaper than having a DS3 ran to an office facility. For example, a lot of companies will purchase a T1 from an ISP for their office connectivity, then colocate their internet-facing servers at the same ISP -- this situation gives their office the up & downstream the need, but gives the servers the enormous amounts of upstream they require.

Consider that when you download a file off the web, you might use 2-3mbps of downstream. Somebody is paying for that upstream remotely so you can download it. The availability of the upstream on the far end is directly related to how much upstream the remote side has sold to it's DSL/Wireless/EverythingElse customers.

 

[spidey07]

Thanks for the info Randal.

So its truly a limitation of the technology - due to carrier frequency? that band being predetermined by ANSI and ITU.

 

[rookie1010]

the limitation is due to bandwidth not tha carrier frequency, right?
i was just thinking about my end as the destination, not thinkin gabout the source of destination, hehe
hey with the ratio of bandwidthd for upstream to downstream being approximately 8:1, does that mean the max DL/UL data rate ratio is 8:1 too?

hey is the modulation, coding the same on ADSL UL & DL?
also what is the FFT sizes?

i got to do the link budgeting for WiMAX-OFDM and was wondering if i could get some of the parameters out of ADSL

also about colocating the webserver, suppose in your example, some one wants to access his companies web facing server, he would do that through the T1 link right, there is no physical direct data link to the colocated server, correct?

 

[randal]

Yea, the technology has ramped up to cope with the allocated frequency space by using frequency division multiplexing. They have OFDM out there (called VDSL for Very fast DSL) that go up to something like 54mbps, but the distance limitations are severe and it isn't standardized yet. The problem is that it runs across unshielded twisted pair and has to coexist with voice - that makes the lower frequencies unavailable (used by voice), and the higher ones unusable (more attenuation/less distance/error prone).

edit: Here in Colorado, Qwest uses HDSL to deliver T1 lines on only two wires. They go T1(CO)->HDSL->T1(NIU)->CPE. They can do this and still guarantee 1.5mbps in both directions because they do not provision the line for voice, but instead utilize the voice freq to carry the larger upstream bandwidth. HDSL is not standardized as far as I know, and Qwest has custom HDSL<->T1 converters. It works out pretty well, but makes me feel like I'm getting raped because I'm paying $200 for a DSL line that costs $28. /edit

Two methods of error correction are generally available and they work together- FEC & interleaving. Forward Error Correction essentially learns where data is likely to be lost, then appends that data to outbound packets (two copies of some data per packet). This increases overhead, but works great ... unfortunately, it falls apart when the broken data is contiguous. Interleaving handles this by scrambling the data in a predefined manner so that if data is lost contiguously on transmission, it's not contiguous loss on reception/de-scrambling -- this allows FEC to work a lot better, because it's then missing little bits here and there, instead of huge chunks.

Depending on the ISP, modulation is DMT or CAP, but 99% of the time it's DMT. DMT basically splits up the the usable bandwidth into 4kHz chunks, then transmits on the ones that do not have interference (determined mathematically via learning algorithm). DMT, FEC & Interleaving are all utilized on both upstream and downstream links.

As for the colo, yes, the "office" side would access the internet-facing server via the T1 line (would not actually cross the internet at large). Most datacenters will aggregate the T1's usage & the server's usage into one bandwidth bill -- that makes traffic from the T1 to the Server free, but traffic from the T1 to the internet at large gets lumped together with the Server-to-Internet traffic, reducing billing headaches & costs. That's what we do, at least

 

[Concillian]

THere are a few services that offer something like 1000 kbps - 2000 kbps up speed for residential prices. They have lots of problems with single users taking all bandwidth by hosting serers, using bit-torrent, etc...

This is why residential policies almost always include no-server usage clauses, and ISPs have been known to cap users who upload excessively.

Also 'typical' users tend to upload a lot less than they download, so if you have, say 768k available it benefits more of your customers to allocate that as 512k down and 256k up than to allocate that 384/384. So customer service is also a concern.

 

[randal]

I am very surprised to hear that any DSL offering -- barring a short-range SDSL delivery -- goes up and over 1mbps upload.

The "768kbps" available is not split into upstream and downstream. The two different streams are each delivered completely independently of each other and have nothing to do with their aggregate service speed. 1.5mbps/896kbps is NOT the same as 1mbps by 1.396mbps or any other combonation that adds up equivalently.

ISPs always purchase their bandwidth full duplex; a 45mbps ds3 is not 22.5up and 22.5 down, it's 45mbps up and 45mbps down. They sell the downstream a lot more than the upstream so that they can 1) limit residential servers and 2) conserve bandwidth for actual server-servers; this *always* results in a disproportionate about of bandwidth being utilized at the upstream side. One of our customer's, a local ISP, consumes ~8mbps of bandwidth downloading, but less than 1mbps uploading. Contrast that with another customer's webserver farm that does ~800kbps downloading and > 10mbps upload. Ratio-wise, our load at our datacenter is split right at about 60:40 download/upload.

 

[spidey07]

Well while we're speaking upstream/downstream.

From a tier 1 provider or any wholeseller - the cost is always for a symetrical pipe. All WAN/MAN technologies that I know of are full-duplex (send/receive at the same time - different channels independant of each other)

 

[Concillian]

Originally posted by: randal
I am very surprised to hear that any DSL offering -- barring a short-range SDSL delivery -- goes up and over 1mbps upload.

I never said the 1mbps offering was DSL.

 

[dmcowen674]

Awesome thread guys :thumbsup:

 

[Goosemaster]

Originally posted by: dmcowen674
Awesome thread guys :thumbsup:

 

[rookie1010]

orward Error Correction does not learn where data is likely to be lost and does not append that data to outbound packets .

it employs block or convolutional decoding properties which correct a certain nuber of errors independently of where they are, as long as they are not conttiguous.

the interleaver takes care of bursty errors, that is errors that occur in bursts or are contiguous.

 

[randal]

http://www.google.com/search?q=define%3Aforward+error+correction

# Adds unique codes to the digital signal at the source so errors can be detected and corrected at the receiver.
www.satnews.com/GLOSSARY.HTML

# A technique used by a receiver for correcting errors incurred in transmission over a communications channel without requiring retransmission of any information by the transmitter; typically involves a convolution of the transmitted bits and the appending of extra bits by both the receiver and transmitter using a common algorithm.
www.e-ratecentral.com/resources/help/glossary/f.asp

# A technique which employs special codes that allow the receiver to detect and correct a limited number of errors without referring to the transmitter. (Carne, 1995)
www.adec.edu/tag/glossary.html

# (FEC) is any technique used to transmit redundancy in a digital signal for the purpose of correcting random bit errors at the receiver without the need for a reverse data link.
www.tw.l-3com.com/tutorial/glossary.html

About the learning part, a more appropriate definition would have been "determines and applies one of several Reed-Solomon coding algorithms to provide redundancy of likely-suspect transmissions to prevent retransmits or implementation of Backward Error Correction" My apologies for the confusion. Regarding FEC, I could be completely wrong so take it with however much salt you please.

 

[rookie1010]

ah, that is what you mean by learning. it uses some sort of adaptive coding techniauqe, and uses a different RS code depending upon the link conditions.