Is there a technical reason that upload data rates are so much lower than download? For instance, we currently get 8Mbps/256Kbps at home, upgraded recently from 2Mbps/256Kbps. That’s 8-32 times faster! And, for our DSL, the upload speed didn’t change with the upgrade…that’s simply what’s offered.
Being a computer guy, I feel like I should know this. But my wife asked me: (1) why there was such a large discrepancy between the rates and (2) why only the download rate could be upgraded. And I couldn’t answer with certainty. Now, I can theorize on some possibilities (that is, totally bullshit):
[ul]
[li]For home endusers, an ISP optimizes for the common case (i.e., they’ll be downloading, not uploading much)[/li][li]Sort of an extension of that (that downloading is the common case), ACKing requires far fewer bytes, thus it’s an architectural design choice[/li][li]An ISP (likely) frowns on a home user running a server; one way to discourage it is to make uploading relatively slow[/li][li]Something to do with CSMA/CD…although I’d have to put some thought into how to bullshit this one ;)[/li][/ul]
But I’d like to know the actual answer, assuming there is one. Any help?
I asked a question to this awhile back. Seems like in Asia the upload/download speeds are equal and it makes a huge difference in p2p applications. When our hotel would do torrent (legit) applications you’d be going along OK then some hotel in Japan would log in and BOOM the torrent would be flying 'cause of his upload rate.
I think the real reason is sales. There’s no huge need for the consumer to upload fast and if you do, then you can get a dedicated line that is equal in up and down speeds. This is A LOT more than the same for a 768 up and the same DL speed.
I think it’s the way our markets in the USA developed (and other European nations) versus how the market developed in Japan, Hong Kong, Singapore, Taiwan and those nations
Right. Depending on the upload rate, there’s a max download rate one can achieve (even when not actually uploading data, due to the ACKs necessary for TCP/IP’s guaranteed delivery). When we lived in IN, I think the d/u rates offered by DSL were the same, although cable’s were different.
Yeah, but I’ve found that for a limitation like this, there’s usually a good technical reason. Perhaps not, but then I’d like to know that’s the case…
Well, there’s only a finite amount of bandwidth available - doesn’t it make sense to allocate it such that most of the pipe is allocated to downloads, which is where most of the data is going?
Thinking about this, I realized I didn’t give your post its due consideration. If the u/d speeds are equal (and, as I understand it, much faster), then that’s evidence that there is no purely technical reason for the difference.
Sure. It makes perfect business sense, just as overselling the total bandwidth an ISP has makes sense. But that’s policy, and I’m curious as to whether there are some vagaries of wiring, or the TCP/IP protocol, or common routing equipment, etc. of which I’m unaware and that might account for it.
The more I think about it, the less I think there is any inherent technical reason for the difference – data communication is (necessarily) two way, and I can’t fathom any necessary cause (outside of policy) for the asymmetry.
Digital Stimulus. You really don’t see that most consumers download way more than they upload? If a line can support 20Mbps what would you rather have 10 up 10 down or 1 up 19 down. I know I prefer 1 up 19 down.
For DSL, the DSL Forum set the standard of how the available frequencies on the copper wires were to be divided between downstream data and upstream data. At the time this standard was initially developed, web surfing was the big app on the internet. If you are web surfing, you need little upstream data because you only need to do the requests for web pages on the upstream bandwidth, the web page which is much larger is carried on the downstream. When manufacturers design their products, they want them to adhere to standards so all of the ASDL DSLAM cards were manufactured to this standard.
In recent years, new standards have been developed that balance out the discrepency between the two somewhat, but the overwhelming demand by residential customers is for the demand on the downstream side.
For people wanting to run servers, there is a different classes of service. It is quickly becoming outdated, but the old standby T1 circuit will provide equal upload and download speeds.
Ah, that’s closer. So, according to your post, it’s just the standards in play. Bandwidth could be divided equally between upload and download capacity, OR either one could be allocated a bigger chunk of the total. It’d be weird (and not very efficient in almost all cases) to have higher upload speeds than download, but it’s possible.
This was bugging me, so I did some more googling and found this WikiAnswers page, which is interesting:
That’s the entire answer given, with no more explanation. Obviously, that answer is only dealing with dial-up modems. And I’d have to bust out my decades-old notes (assuming I can find them) to refresh my knowledge of Shannon, Nyquist, various types of spectrum splitting (I think that’s what it was called), etc. to actually grok it.
As we all know, just 'cuz it appears on teh interwebs doesn’t mean it’s true, but that’s the type of technical answer I’m looking for, if one exists.
I don’t know the direct answer to the OP’s question, but I’ll point out that completely maxing out your upload will decrease your download. I believe the technical reason is that even downloads require a little bit of upload to work, as they have to verify that they got the download.
Right. TCP is “guaranteed delivery”, which means that for every segment of data received, the receiver sends an ACK (acknowledgement) to the sender confirming what data was actually received. Without receiving an ACK within a certain timeout period, the sender will re-transmit the data. UDP, on the other hand, is not guaranteed, and thus has less overhead (and often many lost packets of information).
Thinking about my OP, I realize now that “technical” was really ill-defined. Sorry about that. I suppose I could try again:
Broadband upload data rates are much lower than download rates. Is this due to a theoretical limit (e.g., the Nyquist-Shannon limit? Some fundamental physical limit (e.g., the speed of light over fiber optic cable)? Inherent in networking protocols (e.g., upload overhead is greater than download overhead)? Or is it simply a policy decision (or dictated by accepted standards)?
I now suspect that it’s simply ISP policy and/or some governing body’s standards, but would love a cite that says so…
Most broadband connections use ADSL or ADSL2+. That stands for Asymmetric Digital Subscriber Line. The asymmetric part is the amount of bandwidth dedicated to each direction of transmission. There is a diagram on page 9 of this document showing the relative amounts for upstream and downstream traffic.
There are standards for SDSL also, but because the majority of users would prefer more download capacity, SDSL is not widely available (in the US anyway).
The “A” in ADSL stands for “asymmetric”; i.e. different upload speed than download speed.
I have noticed that in the US, it is usually just referred to as “DSL” whereas over here in the antipodes it is commonly “ADSL”. Or the annoyingly ambiguous “broadband”.
SDSL was (is) available most often to businesses in the US, where data transfers are often large-volume and bi-directional.
As to one technical reason why SDSL isn’t common in the home, high quality line drivers are typically used on the transmit side for higher speed, while the receiver circuits are pretty simple. It’s cheaper to run ADSL to residences as the high quality transmitters and expensive equipment are at the providers switch site (whether central office or remote), and the users home modems are extremely inexpensive.
The other reason for asynch data rates literally is demand, as mentioned above. More residences spend time DOWNLOADing, be it video, music, web pages, whatever. By providing the bigger pipe on the download side, customers don’t even notice any slowness on the upload side.
OK, there is a reason why Asymmetric DSL is so popular. There is a key idea as to why it is used. It is about interference and the relative signal to noise ratios that can be managed.
So - as alluded to above - Shannon showed that there is a maximum information transfer rate possible in any channel given its bandwidth and signal to noise ratio. (DSL and ADSL divide the available bandwidth up into a large number of seperate mini-channels so you really need to work out the transfer rate possible with each one of these and add them all together - but this additional complexity doesn’t actually change the core issue here.)
So, you have bandwidth - this is limited in the end by the wire between you and the DSLAM (the gear in the exchange.) The longer the wire the lower the bandwidth becomes. And you have your signal to noise ratio. This measures how much other junk you see on your wire versus the signal. The longer the wire the more loss you get, so the signal drops, and the worse the quality of the wire the worse it gets too. Noise can be thermal noise, electrical interference, but for DSL lines the critical source is traffic on other wires that share the same bundle as you do. Once we know the bandwidth and the signal to noise we know from first principles the absolute maximum information transfer rate possible. So far so good.
Now you talk to the DSALM along a single wire pair. So signals coming compete with signals going. The intuition would be that you can therefore partition this whatever way you want. Take the available capacity and aportion it in any useful combination. But there is a catch.
We can note that the situation with you and the exchange isn’t symmetric. You are out at the end of a long wire, and nearby you will have other neighbours that share the same cable bundle, but not all that many, and in general things are not all that crowded near you. However at the exchange end the DSLAM has every single subscriber connected to it, and it is sending lots of data down all those cables, all of which are closely bundled up. The DSLAM end of the cable sees a lot more interference than your end. When it sends you data it sends the signal at a nice high level, and that signal travels down the wire, losing energy all the way to you. The interference on the wire from things at the DSLAM end is also attenuated at the same rate. Since your end of the cable is pretty quiet, with only a few neighbours around, the locally picked up interference isn’t all that great. But going the other way, you send a signal back to the DSLAM, it gets attenuated the same amount on the way back - but the huge crowd of interfering nearby cables at the DSLAM end means that there is much more interference for the DSLAM to contend with when listening for your data. The upshot is that it is that the DSLAM sees a worse signal to noise ratio on the line than you do, and thus you can never send it data as fast as it can send to you.
You can trade off the upload and download bandwidth’s but it isn’t a one to one tradeoff. There is a varient on ADSL supported by some carriers (Annex-M) that allows you to double your upload speed from 1Mb/s to 2Mb/s. However your download speed will drop by more than 3Mb/s to compensate. Fully symmetric DSL is worse still - the price for the symmetry is an aggregate bandwidth that is even less.
The chosen partitioning of bandwidths for ADSL are a happy mix of providing a good partitioning of trasnfer rates in a way that suites most home users, and one that turns out to be even more advantageous (getting much more download in return for upload) than you might first think due to the asymmetry in the nature of the communications.
And the reason those asian countries have symmetric upload/download is because they are not using (A)DSL. Their comms infrastructure is more recent, and the Korean Govt has aggressively pushed the telecoms companies to use fibre to the home, and separate IP networks (rather than circuit-switched POTS).
In many cases, subscribers are connecting to the internet by ethernet over very high bandwidth links. Thus, they are looking at 1Gb connections in the very near future.