Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[kezzaman]

Hi guys,

Wondered, if there's alot of bitswapping on a line does it indicate that there might be an issue?
and does all that swapping affect latency, like are packets having to wait while bits are being swapped, or am i just not getting what bitswapping is about? 

[burakkucat]

. . . or am i just not getting what bitswapping is about? 

Possibly. 

Kitz has a very nice write-up on the subject.

[snadge]

just too add. no it wont affect latency... 

as you talk of latency I suspect your a gamer? Gaming latency relies on many small packets (per second) being sent & received, this uses a fraction of your available bandwidth.

# think of your broadband connection like a multi-lane highway (many hundreds of lanes), with broadband speed (throughput) its how many cars you can get down that highway in X amount of time, dodging traffic and roadworks on the way, you can send many hundreds of cars all behind each other until all your available lanes are full, and however many get down that highway in X amount of time is how fast it is... now with Latency its how quick one single car get down the highway and back on its own - so when your gaming its like a small convoy of cars all running in a line one after the other down just a couple lanes on this wide open highway, if traffic becomes too much on this highway then it can affect each cars finish time (response time) , where as how fast you can download is like sending every car you have until all lanes are full - i think thats a good metaphor... - full download speed is congesting your highway to get as many cars through as poss and latency is just a small trickle of cars going through.

Anyway, like I say Bit-Swapping is normal and will not affect your latency, well, unless you had a very long line with a really slow & noisey connection and the bits constantly being swapped consisted of a large percentage of all available bits/tones, and the router cpu was slow at swapping them?? ..perhaps anyway, maybe someone can clarify that? - but as far as Im aware it will not

[kitz]

Good analogy there snadge with the car lanes

Latency is how long it takes a single packet of data to reach from point A to point B.  Latency should remain the same regardless of your adsl speed.  eg someone on a 512kbps connection could very well have exactly the same latency as someone on a 24Mbps connection.

adsl works by opening up lots of different channels (lanes) and allowing more data packets to traverse at the same time which is what increases the bandwidth speed.  Think of a hosepipe.. it takes the same time for a drop of water (data packet) to reach the tap to the end of the hose, but if you have a wider pipe then more water (datapackets) can pass through the hose at the same time.

>> all that swapping affect latency,

No, bit swapping makes sure that the maximum databits can pass through on each channel (lane) and if one channel cant handle x no of bits it swaps them to another channel if that has any spare.  kinda like lane hopping..  except that your router is constantly monitoring the bit allocation and is in advance thinking which traffic (bits) to send down which lanes (channels).   It would be more like a policeman directing traffic to the best route.

Interleaving and error correction can affect latency though.   This is because of 1) the slight extra time whilst data is interleaved by the router and reassembled at the other end  2) Error correction carries redundant data - therefore less useful data per transmit (speed)..  and of course the time taken by the routers at both ends to code/decode the data.

>> unless you had a very long line with a really slow & noisey connection and the bits constantly being swapped consisted of a large percentage of all available bits/tones, and the router cpu was slow at swapping them?? .... maybe someone can clarify that

Correct, bit swapping only works to a certain extent.. if the line becomes too noisy, then theres not enough 'spare' for bits to be swapped to... and data may have to be retransmitted...  and yes lots of bitswapping can cause a router cpu to work hard.

[kezzaman]

Thx guys 

[kezzaman]

Could you say that these highway lanes are tones?

[burakkucat]

Yes. 

The tones (as in DMT) or discrete channels. 

[kezzaman]

Cool 

Lets say that on an adsl2+ line (with tones going from 33-511) i was downloading a file which was only downloading at half the possible speed, and lets assume that each tone is equal in size (same amount of bits). would this download use half the tones, for example 33-272, or would it use half of each tone, for example all the tones 33-511 but only using half of each 1?

[kitz]

ummmm  good question.


Ive no idea which tones it would use.  DMT just makes the tones available for use if needed.  It opens up the lanes on the highways with a maximum speed limit but doesnt dictate which lane the traffic will use.

I suppose theres many instances where traffic doesnt traverse at its full potential say for example you have network cable and a network switch that is gigabit ethernet..  but you are restricted to 10/100 because of the NIC.

[burakkucat]

Ooh. A good question. But one that I do not know the answer. 

I best keep watch . . . ready to learn.

[kitz]

>> I best keep watch . . . ready to learn

lol me too..   

I really dont know..  but to me, logic would seem to point to it filling the bins in order... starting at the lower end of frequencies until each bin is full.   
I would imagine the router having to work out and constantly monitor the existing throughput speed, and divide it up between the total no of bits available... and then start calculating say 2 bits per bin then a fraction of a second later 3 bits per bin would add unnecessary additional work for the router.

[kitz]

Still none the wiser despite googling...  it would appear there are several different methods..  some of the tech stuff is over my head... but it would seem to imply each bin is analysed in order.

However whats the waterfill approach?  anyone know what that means?

-----------
edited to add

>>  whats the waterfill approach?

I think it means filling the lower (better) channels first...  or at least it does when talking about power allocation.

[kezzaman]

If i had to guess i would think that filling each bin in order might create issues if theres alot of bitswapping going on. For example, if the 1st couple of bins are full but bits need to be swapped around in them it could create abit of lag perhaps.

>>> whats the waterfill approach?

But this "waterfill" suggests to me that yea each bin is filled up and then when its starts overflowing it goes into the next bin, 

... Unless each bin has tiny little holes in that the water is allowed to seep through, filling each bin equally. 

L0L

[kezzaman]

>>  whats the waterfill approach?

I think it means filling the lower (better) channels first...  or at least it does when talking about power allocation.

I think i get that, imagine turning the DMT graph upside down so the more stable bins become deeper channels.

[kitz]

>>> but bits need to be swapped around in them

bits can be swapped from any subchannel in the whole frequency range, not just adajcent. When the bit table is allocated theres normally some spare SNR reserved for the bit-swapping process.


By filling the early (low frequency) bins first...

1) Usually these are the more stable channels.
2) if the line is only running at half speed... there will be plenty of empty bins doing nothing in the lower end where bits can be swapped from.
3) if the line is running at full pelt all bins are full so it will utilise bins with any spare SNR.  At this point power output can be increased to gain a bit more SNR. (real SNR not SNRM)

Bit swapping happens really fast.. and 1 bit of data will pass through so quick, that it will be 'gone' and dealt with in less than the blink of an eye anyhow.

Thinking about it, if you need to take bits from bins.. then IMHO it would be better if possible to  'take' from the lower end (higher frequencies) where things are likely to be less stable anyhow, rather than borrow from the higher frequencies.

Waiting to see if asbokid (or anyone else) can add anything.  He may be able to make sense of some pretty complicated algorithms.. cause thats where I tend to get lost. :/

>>> imagine turning the DMT graph upside down

ah... yes I see that