bandwidth, data rate and its upgrade

antennaboy35

I think you are right, but your explanation is pretty clear and simple....I don't see anything too complicated though.
As you are saying, packets just arrive. The computer that is downloading more stuff is probably requesting packets faster and keeps the router busy .....
I am happy with that general understanding for now....Trust me, it helps a lot...
My questions are very basic, I know: I spend some time in coffee houses and I see all these computers surfin the web, downloading, all at the same time. Packets travels around, with their own addresses and destination...It is an amazing process going on. And for me, being still a little uneducated on the topic, my questions are a little kiddish.
thanks
antennaboy

Forsaken_GA

drkat wrote: »

Typically a cable modem will be pushed a new bin file and it tells the device at what speed to operate at.

The carrier network isnt limited but the device is provisioned or the port is limited.

With the cable modem, it depends entirely on the speeds you're talking about, and what docsis standard the modem and the CMTS supports. Docsis1/1.1/2 support a max of 42 megs down (in the US), period, because it only uses a single channel. So for speeds up to that, then yes, the data rate can be adjusted simply by pushing a new config file.

In order to get above 42 megs, the CMTS and the cable modem has to support DOCSIS3, and the speed increases are gained through channel bonding (this is also a primary driver on the parts of the cable companies to get rid of analog channels... the more analog channels they reclaim, the more channels are available to use for downstream traffic, and digital channels take up far less frequency than analog channels)

Forsaken_GA

antennaboy35 wrote: »

thank you (thanks everyone). I ask for patience, soon my questions will be more mature and advanced. Sometimes the basics are insidious....

So packets arrive at the router from multiple computers. No round-robin but FIFO method. Packets are put in a queue and processed, one at a time...

How do I interpret what is going on when there are 2 computers, computer A and computer B, wirelessly connected to a router in a HAN and computer A has engaged into heavier downloading, like videoconferencing or else, and it is sucking all the bandwidth?

How does A manage to get control of all the bandwidth? Is it throwing packets at the router faster than B? Both A and B are sending packets to the router, but why does the one involved in heavy downloading manage to keep the control of most of the bandwidth?

What if B (slow activity, few requests) started its browsing activity before computer A started its heavy downloading? Would the bandwidth still go to computer A?

Ok, I'm going to water this down quite a bit. Assuming you're talking TCP traffic, what happens is that Host A saturates the pipe. So when Host B sends out some traffic over it's session, because Host A is monopolizing the traffic, the return traffic from Host B's request is lost. Well, when TCP detects dropped packets, it lowers it's window size and then goes through retransmit. This results in lower throughput. When that happens enough time, one side of the connection usually just says screw this and it drops the connection. Since the heavy usage host is monopolizing all the bandwidth, other hosts never really get a chance to get started, and those that do will receive very low data rates because their window size is so small.

It gets even more fun when you throw UDP traffic into it, because it has no concept of window size and connection, so both sides will just keep tossing out packets and dealing with the loss. This is why heavy UDP usage can choke out TCP flows pretty easily over a period of time.

I'd recommend you pick up a copy of Doug Comer's Internetworking with TCP/IP, understanding TCP/IP interaction is core to understanding IP data flow

Forsaken_GA

antennaboy35 wrote: »

I am reading this book that talks about wired routers speeds. The book says that they range from 100 Mbps to 1000Mbps. but there are also high performance routers that go 10 Gbps....
what is the point of that speed? I don't know of any internet connection that fast?
Is that 10 Gbps speed useful for an internal network, between computers in that network? A modern computer can output data at Gbps speed correct?

I think folks may have gotten a little overtechnical on their explanations of this.

The point of links that fast is to carry a whole lot of traffic. Those kinds of links are normally used for backbone and transit links. They're not provided to end users.

You need to understand the concept of backhaul traffic. From a traffic flow perspective, this is what happens when you type techexams.net into the browser at the top (skipping all the dns and such)

Your packets get sent to your default gateway. On a consumer residence using a router, this will be the home router. The home router then sends it to it's default gateway, which will be whatever device terminates the other end of the connection... a CMTS in a headend, a DSLAM in a CO, whatever, it'll be your providers router.

From there, it doesn't get routed directly to the internet. Each provider has different points of connection to the internet, usually in major cities. For example, Atlanta, Blackburn, Seattle, Dallas-Fortworth, etc. This is called backhaul, because the traffic is being backhauled from it's collection point to a central distribution point for routing to it's end device.

So lets say you're in Miami, FL, and your ISP's distribution point is in Atlanta, GA. The sp's terminating device sends the packet on up there, and it does this for *every* single one of it's customers that terminate to that device. That's why you need the high speed links. If you've got 10,000 customers off that terminating device, and each uses 1 meg, then you need to backhaul 10 gigs of data up to the distribution point so it can go out to the public internet. Now you could run a whole bunch of smaller links, or you could drop all of that into one pipe. The big fat pipe tends to be more cost effective.

antennaboy35

Thanks Forsaken_GA, I just went to the library and got a copy of Doug Comer's Internetworking with TCP/IP....
I will read it up and get back to you...

thanks
antennaboy

antennaboy35

From Comer's book, I get that when we pay for data rate to be delivered to our home the ISP gives "priority" to some traffic (e.g., customers who pay more), the ISP must insert a device into their network that handles traffic engineering: maintain queues of packets and select packets from some queues more frequently than from others.

Page 224: Packet switching requires a sender to divide each message into packets. Statistical multiplexing occurs among packets rather than among bits or bytes. Once a sender gains access to the underlying channel, the send transmits an entire packet, and then allows other senders to transmit a packet. When no other senders are ready to transmit a packet, a single sender can transmit repeatedly.

it_consultant

vinbuck wrote: »

There are actually some advanced bandwidth controls you can put in and rate-limit just about wherever you want to in order to meet certain requirements. Rate limiting can be done at the port or on certain types of traffic - combine that with QoS and you can pretty much dictate what each end "packet generator" will and won't do. The only major limitations are hardware capability and administrative overhead. However, most Layer2/3 switches deployed these days support a wide variety of rate-limiting options.

It seems to be more common in the Enterprise to let the end user/device have the full bandwidth of the access port because it is easier but in the carrier world, rate-limiting is what we specialize in and there are some neat tricks you can do with rate-limiting to smooth out the network.

Rate limiting a port is different than dividing available bandwidth evenly. In the ISP world I imagine rate limiting happens frequently, the question, as I understood it, would be the other way around. We already have a rate limited port of say, 10 MB, the devices who are generating traffic will not have an equally divided share of the bandwidth unless someone configured each of their switch ports to be rate limited at the exact denominator which when aggregated, would equal 10 MB. The main flaw with this is that it would rate limit access to LAN resources as well.

As far as I know, most QOS tools wont equally divide bandwidth either, they simply tag some traffic higher priority than other traffic.

it_consultant

Forsaken_GA wrote: »

I think folks may have gotten a little overtechnical on their explanations of this.

The point of links that fast is to carry a whole lot of traffic. Those kinds of links are normally used for backbone and transit links. They're not provided to end users.

You need to understand the concept of backhaul traffic. From a traffic flow perspective, this is what happens when you type techexams.net into the browser at the top (skipping all the dns and such)

Your packets get sent to your default gateway. On a consumer residence using a router, this will be the home router. The home router then sends it to it's default gateway, which will be whatever device terminates the other end of the connection... a CMTS in a headend, a DSLAM in a CO, whatever, it'll be your providers router.

From there, it doesn't get routed directly to the internet. Each provider has different points of connection to the internet, usually in major cities. For example, Atlanta, Blackburn, Seattle, Dallas-Fortworth, etc. This is called backhaul, because the traffic is being backhauled from it's collection point to a central distribution point for routing to it's end device.

So lets say you're in Miami, FL, and your ISP's distribution point is in Atlanta, GA. The sp's terminating device sends the packet on up there, and it does this for *every* single one of it's customers that terminate to that device. That's why you need the high speed links. If you've got 10,000 customers off that terminating device, and each uses 1 meg, then you need to backhaul 10 gigs of data up to the distribution point so it can go out to the public internet. Now you could run a whole bunch of smaller links, or you could drop all of that into one pipe. The big fat pipe tends to be more cost effective.

What he is describing is an ISP level "handoff". Forethought.net hands off traffic to Qwest, Level 3, Verizon, ATT etc. Forethought.net is my ISP. Most of the traffic generated inside of Forethought's network will be bound for one of these high level providers which plug into most major website directly AND will connect you to downlevel ISPs across the country or internationally. Therefore their equipment must be able to not only handle a great deal of traffic but also route it efficiently. You *could* send a packet destined for Time Warner over ATT, but it is more efficient to use Level 3.

Forsaken_GA

it_consultant wrote: »

What he is describing is an ISP level "handoff". Forethought.net hands off traffic to Qwest, Level 3, Verizon, ATT etc. Forethought.net is my ISP. Most of the traffic generated inside of Forethought's network will be bound for one of these high level providers which plug into most major website directly AND will connect you to downlevel ISPs across the country or internationally. Therefore their equipment must be able to not only handle a great deal of traffic but also route it efficiently. You *could* send a packet destined for Time Warner over ATT, but it is more efficient to use Level 3.

Well, no, I hadn't even gotten to the hand off point yet, it wasn't necessary to explain it that far, as needing lots of bandwidth on the backhaul link serves as a demonstration of the need of high capacity links before the traffic ever leaves the providers network. There's a whole lot of routing that goes on inside a providers own network before it ever makes it to an egress point.

As far as efficiency goes... that all depends on your peering arrangements. If Time Warner was directly peered with AT&T (which they are), it's more efficient to send it directly to them than to send it out over Level3, especially if that peering between AT&T and TWC is settlement free, but that's more about the politics of Interdomain routing than the technology

antennaboy35

today I was told that a cable CMTS or DSL head-end modem has a built-in rate limiting function, known as a "traffic shaper". All the packets headed for your house are placed in a queue, and released at whatever rate is in your service agreement. There's a timer that runs an algorithm known as "leaky bucket" (i.e., it "drips" packets to you). The subscriber's rate is a parameter to the leaky bucket algorithm that can be reset easily when you pay more.....

Any correction or comment on this?

antennaboy35

I asked a professor of mine the following questions:

--Say we have 3 computers, A, B, C in a home, all 3 connected wirelessly to the internet via the wireless router/switch. We know from experience that each computer sends and receives packets to the router. But if computer A is engaged into heavy downloading it will also dominate the whole bandwidth and the other computers will be slow... Why? Is computer A continuously exchanging packets and keeping the router busy in such a way that packets from the other computers are processed slowly?


The Wi-Fi protocol uses a probabilistic algorithm, so it is *possible* for one computer to dominate. However, if all the network interfaces are built correctly, chances are high that each will receive a fair share.

--If computer B and C started surfing the web before A started its downloading session, would A still take up most of the bandwidth?


On the other side of the router (back at the ISP), there is a packet queue. If Computer A is downloading, it has probably advertised a large buffer
(i.e., a TCP window). Across the Internet, the sender will send enough packets to fill the window. When all those packets reach your ISP, they
go into the queue for your house. Now, if someone browses the web, the web server will also send packets, but they will go to the back of the
ISP queue. So, it may *appear* that computer A is hogging all the wireless bandwidth, but in fact it's probably just a result of the buffering.

Any comment or further clarification on this?

thanks
antennaboy

it_consultant

antennaboy35 wrote: »

I asked a professor of mine the following questions:

--Say we have 3 computers, A, B, C in a home, all 3 connected wirelessly to the internet via the wireless router/switch. We know from experience that each computer sends and receives packets to the router. But if computer A is engaged into heavy downloading it will also dominate the whole bandwidth and the other computers will be slow... Why? Is computer A continuously exchanging packets and keeping the router busy in such a way that packets from the other computers are processed slowly?


The Wi-Fi protocol uses a probabilistic algorithm, so it is *possible* for one computer to dominate. However, if all the network interfaces are built correctly, chances are high that each will receive a fair share.

--If computer B and C started surfing the web before A started its downloading session, would A still take up most of the bandwidth?


On the other side of the router (back at the ISP), there is a packet queue. If Computer A is downloading, it has probably advertised a large buffer
(i.e., a TCP window). Across the Internet, the sender will send enough packets to fill the window. When all those packets reach your ISP, they
go into the queue for your house. Now, if someone browses the web, the web server will also send packets, but they will go to the back of the
ISP queue. So, it may *appear* that computer A is hogging all the wireless bandwidth, but in fact it's probably just a result of the buffering.

Any comment or further clarification on this?

thanks
antennaboy

The router isn't slowed it is the connection speed which is hogged by the chatty computer. You could fill the memory of your router if you have a cheap router and a lot of connections, say in a scenario where you are torrenting. However, I can just as easily take up all the available bandwidth with an FTP session which is only one connection and is boring to the router.

It seems like you are overthinking this. If you have a 20 MB pipe and one computer is taking almost all of that, your other computers simply have less to use than they normally would, ergo their connections are slower. Most home bandwidth is a mere fraction of advertised speeds over a connection period of more than a minute, so this behavior is noticed obviously. On an enterprise connection we will regularly see 75% of available bandwidth taken before web browsing is affected.

Forsaken_GA

antennaboy35 wrote: »

today I was told that a cable CMTS or DSL head-end modem has a built-in rate limiting function, known as a "traffic shaper". All the packets headed for your house are placed in a queue, and released at whatever rate is in your service agreement. There's a timer that runs an algorithm known as "leaky bucket" (i.e., it "drips" packets to you). The subscriber's rate is a parameter to the leaky bucket algorithm that can be reset easily when you pay more.....

Any correction or comment on this?

That's not an absolute, but yes, virtually all modern equipment has some form of rate limiting available, though the industry appropriate term is QoS (Quality of Service). Your maximum rates are going to be determined entirely by what technology is in use, but the ISP has the capability to limit your actual throughput to any point below that, depending on what you've subscribed to.

antennaboy35

thanks it_consultant.

You are probably right, I am overthinking it. But in technical terms, when you say "...that one computer is taking almost all of that," do you mean that it is flooding the router with packets?
Usually, when a computer sends a request for some content in the form of packets, to a server computer that is hosting the content itself. The server then sends the content to the requesting computer, in the form of packets....
Does the flow of those packets, associated with that specific large content, sort of clog the router? Why does it get priority over the other computers sending and receiving their own parckets?
thanks
antennaboy

networker050184

antennaboy35 wrote: »

thanks it_consultant.

You are probably right, I am overthinking it. But in technical terms, when you say "...that one computer is taking almost all of that," do you mean that it is flooding the router with packets?
Usually, when a computer sends a request for some content in the form of packets, to a server computer that is hosting the content itself. The server then sends the content to the requesting computer, in the form of packets....
Does the flow of those packets, associated with that specific large content, sort of clog the router? Why does it get priority over the other computers sending and receiving their own parckets?
thanks
antennaboy

Same as earlier, the router just processes what it gets. So if one server is sending more traffic than another its going to take more of the bandwidth. There is no priority by default, first come first serve.

Forsaken_GA

antennaboy35 wrote: »

thanks it_consultant.

You are probably right, I am overthinking it. But in technical terms, when you say "...that one computer is taking almost all of that," do you mean that it is flooding the router with packets?
Usually, when a computer sends a request for some content in the form of packets, to a server computer that is hosting the content itself. The server then sends the content to the requesting computer, in the form of packets....
Does the flow of those packets, associated with that specific large content, sort of clog the router? Why does it get priority over the other computers sending and receiving their own parckets?
thanks
antennaboy

It doesn't really get priority, it's just acting like a bully. This goes back to how TCP/IP works, and why the need for QoS was developed.

So you get one computer that starts a large transfer while the pipe is quiet. Since it has all this bandwidth available, it has a large transmit window and keeps using it because it can.

So when another computer decides to start a transfer, the tcp session loses packets, because the one that has a session already in progress is taking up all the bandwidth. Since the second transfer has lost packets, the rules of TCP says to slow down and cut the window size in half and then retransmit. It will keep doing that until either one side of the connection gives up for good, or until it finds a window size small enough that it's not losing packets, which will usually be at a much slower speed.

There's only two ways around this:

#1 - Have a bigger pipe than the other side has. If I have a 10 gig pipe, and the guy I'm transferring from only has a 1 gig pipe, then I can't possibly fill the pipe on that one transfer, so there's bandwidth left for the other computers

#2 - The gear that's routing/switching the packets enforces some method of rate limiting, whether that's a hard limit in transfer rate, or traffic shaping, etc that will make individual nodes behave.


There's no priority until you get into QoS, where actually can assign priority, it's just the way TCP/IP works. The first guy to get there gets whatever he can get his hands on, and by default, there are no rules in place to that first guy play nice with others on the same link.

antennaboy35

Ok, thanks everyone, I am pretty happy with my understanding, thanks to you.....

By the way, is Gigaethernet popular and available these days? It takes care of the bottleneck in the last mile, correct?
Who gets something like that? A small/large company? Does it have distance limitations?

Ethernet is a protocol. The wiring used to be twisted pair but I guess it can be delivered via fiber, coax, wireless, correct?

As far as internet connections for home, DSL and cable are the most popular and fastest ,correct? Or is there something else I am not aware of?

Do all connections start at someone home and the first stop is at the central office CO or not necessarily? Is the central office always involved?
The central office always belong to the ILEC. Do CLEC have their equipment there too?

thanks,
antennaboy