Solved: ATM concepts ..... need clarification

Manoj Wadhwa · ‎03-16-2008

Hi Friends,

I understand this is a simple question considering the quality of this forum, but i need you help to clarify on some of the concepts of ATM.

I understand ATM is a Layer2 Technology which has many benefits over other technologies like LL, FR etc. What i want to know is how is ATM terminate on a router. Does the router need a separate ATM module to support it? Is the ATM/ IMA interface on a router a physical or a logical interface. Any good links that give some good description of this technology would be greatly appreciated. Thanks in advance.

Regards,

Manoj

marikakis · ‎03-16-2008

Hello Manoj,

I laughed out loud when I read this last post of yours, because although you claim to be confused, it seems to me that you got straight into the point of this whole ATM story and even more. Tough questions you put into the table! :-)

1. In most cases the answer is yes. Inside the provider network SAR is done at their own routers as well for their own point-to-point backbone VCs (that have nothing to do with direct customer physical connectivity).

2. The networking history is full of stories of the pretty-cool-new-technology that takes over the fading-away-legacy-technology. ATM has been in both of these categories. Besides the economical interests of networking device companies to keep pushing new things for us to buy and the intellectual-economical interests of researchers to justify their existance, sometimes those new technologies do have something to offer. Not that much that the marketing claims, but a few new things are there. From a technological standpoint in many cases it seems like going back-and-forth or in circles. MPLS was supposed to be fast, but nobody puts it in their network for this reason. The providers for example, see MPLS as an opportunity to sell new services. The driving factors for new things are supposed to be cost, speed, scalability to name a few. Yet, nothing new comes without a cost. High speed networking and high performance anything is full of comprimises.

3. ATM is L2 technology. There are various options for its L1. One common option is optical interface on the ATM module card where 2 optical cables are attached (trasmit/receive):

http://www.cisco.com/en/US/docs/interfaces_modules/port_adapters/install_upgrade/atm/pa-a3_ATM_install_config/5117ovr.html#wp1038892

Another cabling option is coaxial with BNC connectors : http://www.cisco.com/en/US/docs/interfaces_modules/port_adapters/install_upgrade/atm/pa-a3_ATM_install_config/5117ovr.html#wp1038876

4. ADSL is L1 technology. To speed up deployments, ATM was specified as the L2 used over ADSL. (For Ethernet both L1 and L2 specifications exist.). The PPPoE, PPPoA and friends are there for the logical termination of the customer connection. While you physical ADSL connection terminates on a DSL port of a card in a DSLAM, your logical connection goes further in the network to get IP address and other information and terminates logically somewhere else. By the way, your home telephone cable is another type of cable that can be attached in an ATM interface (since ATM is L2 for ADSL, but others could be as well).

I hope you did not really mean that this was just a starter. :-)

Kind Regards,

M.

View solution in original post

mohammedmahmoud · ‎03-16-2008

Manoj,

Agree with Maria, i LOL my self :), but its very nice to meet people looking into such details, i mean having an engineering point of view.

1. SAR should be handled at the device responsible to send non-ATM traffic out of its ATM interface as ATM traffic, which can be a CPE or a provider Core router with an ATM interface, it is an essential process for any ATM device and usually done by a dedicated chip (SAR engine).

2. To add another comment to Maria's, ATM suffers design complexity in the hardware design, for example faster SAR engines are more complex and expensive, which causes the ATM technology to be speed limited and expensive as a WAN technology for SP backbones, plus the complexity of managing devices of multiple various technologies.

3. The most common layer 1 technology used with ATM is SONET/SDH (fiber interfaces) as illustrated by Maria.

4. The main factor that has driven ADSL to adopt the use of ATM was the ability to provision multiple PVCs on the same physical link (which can't be done via PPP alone), more over the invention of PPPoA and PPPoE solved the issue of requiring PPP (with all its addons) over such implementations.

BR,

Mohammed Mahmoud.

View solution in original post

mohammedmahmoud · ‎03-16-2008

Hi Manoj,

You are very welcomed. Any layer 2 Technology requires a relevant physical interface, Ethernet requires an Ethernet interface, FR and PPP requires a serial interface, and accordingly ATM requires an ATM interface, keeping it simple, the ATM interface will not be responsible for the layer 2 encapsulation only but it also does a job between layer 2 and layer 1, baring in mind that also different layer 2 technologies use different physical interfaces (ex: Ethernet using RJ-45, while FR using V.35 for example).

Further on, ATM is completely different than other packet switched technologies. As a start you can go for wikipedia, further you can go to deep technical papers:

http://en.wikipedia.org/wiki/Asynchronous_Transfer_Mode

http://www.cisco.com/univercd/cc/td/doc/cisintwk/ito_doc/atm.htm

http://www.cisco.com/warp/public/614/12.html

BR,

Mohammed Mahmoud.

marikakis · ‎03-16-2008

Hello,

When you read about ATM, it seems like a big deal, until you make the pieces work together with your own hands.

From what you have been reading (and what you will read in the URLs already posted), the most important in understanding how ATM network functions is the concept of "ATM cells". The ATM network switches fixed-size (53-byte) cells. But hey, in other technologies, when we where to forward an IP packet, we where just encapsulating the IP packet according to the L2 technology of the outgoing link. How do we feed the ATM network with the cells it requires? We take the IP packet, put it inside ATM Adaption Layer (AAL) header/trailer to construct a "frame" and then chop the "frame" into cells.

So, imagine a network with ATM switches that only know about switching cells (know nothing about IP). You connect your router to this network via a physical router interface that sends valid (validity is determined by ATM VC setup) cells into the network. This interface should also be capable of reconstructing an IP packet from its component cells (normally more than one cell, cell is only 53-bytes). The creation of cells and the collection of those cells later is a process called SAR (segmentation and reassembly). The ATM interface of edge devices of the ATM network needs a SAR module.

Now imagine the traffic flow. Your edge device receives an IP packet from an Ethernet interface, strips the Ethernet information and decides to send IP packet via a VC that has been set up over an ATM interface. The IP packet is dressed inside the AAL (AAL5 is very common), chopped and sent over the ATM network. The cells reach another edge device ATM interface that collects the cells and reconstructs original IP packet. If IP packet must now be forwarded over an Ethernet interface, we do have our IP packet and proceed as usual.

I have only mentioned (not described) the concept of VCs and how ATM network switching is done. If you need more on those, please say so.

Kind Regards,

M.

Manoj Wadhwa · ‎03-16-2008

Hi Mohammed & Marikasis,

First of all, thanks for your replies. You are right, indeed i was going through the links and it just got me think how complex this technology is. However, with your explanations, it has made be feel better now. But i still have some queries and i need your help to clarify further.

1. From my understanding now, the SAR is done by the CPE router and not by the Service provider, right

2. If ATM is still so fast, why is it fading away. What are the common environments where using ATM's is better than other technologies like FR, MPLS etc.

3. I am curious to know how an ATM interface looks like. Searching through google does not give much results, at least to me :-)

4. I have also been going through ADSL. Why does ADSL use ATM as its layer 2. PPPoE, PPPoA just adds salt to my confusion.

I know its a lot more ... but as a starter, i hope you understand my queries :-)..

Thanks again.

Best Regards,

Manoj

marikakis · ‎03-16-2008

Hello Manoj,

I laughed out loud when I read this last post of yours, because although you claim to be confused, it seems to me that you got straight into the point of this whole ATM story and even more. Tough questions you put into the table! :-)

1. In most cases the answer is yes. Inside the provider network SAR is done at their own routers as well for their own point-to-point backbone VCs (that have nothing to do with direct customer physical connectivity).

2. The networking history is full of stories of the pretty-cool-new-technology that takes over the fading-away-legacy-technology. ATM has been in both of these categories. Besides the economical interests of networking device companies to keep pushing new things for us to buy and the intellectual-economical interests of researchers to justify their existance, sometimes those new technologies do have something to offer. Not that much that the marketing claims, but a few new things are there. From a technological standpoint in many cases it seems like going back-and-forth or in circles. MPLS was supposed to be fast, but nobody puts it in their network for this reason. The providers for example, see MPLS as an opportunity to sell new services. The driving factors for new things are supposed to be cost, speed, scalability to name a few. Yet, nothing new comes without a cost. High speed networking and high performance anything is full of comprimises.

3. ATM is L2 technology. There are various options for its L1. One common option is optical interface on the ATM module card where 2 optical cables are attached (trasmit/receive):

http://www.cisco.com/en/US/docs/interfaces_modules/port_adapters/install_upgrade/atm/pa-a3_ATM_install_config/5117ovr.html#wp1038892

Another cabling option is coaxial with BNC connectors : http://www.cisco.com/en/US/docs/interfaces_modules/port_adapters/install_upgrade/atm/pa-a3_ATM_install_config/5117ovr.html#wp1038876

4. ADSL is L1 technology. To speed up deployments, ATM was specified as the L2 used over ADSL. (For Ethernet both L1 and L2 specifications exist.). The PPPoE, PPPoA and friends are there for the logical termination of the customer connection. While you physical ADSL connection terminates on a DSL port of a card in a DSLAM, your logical connection goes further in the network to get IP address and other information and terminates logically somewhere else. By the way, your home telephone cable is another type of cable that can be attached in an ATM interface (since ATM is L2 for ADSL, but others could be as well).

I hope you did not really mean that this was just a starter. :-)

Kind Regards,

M.

mohammedmahmoud · ‎03-16-2008

Manoj,

Agree with Maria, i LOL my self :), but its very nice to meet people looking into such details, i mean having an engineering point of view.

1. SAR should be handled at the device responsible to send non-ATM traffic out of its ATM interface as ATM traffic, which can be a CPE or a provider Core router with an ATM interface, it is an essential process for any ATM device and usually done by a dedicated chip (SAR engine).

2. To add another comment to Maria's, ATM suffers design complexity in the hardware design, for example faster SAR engines are more complex and expensive, which causes the ATM technology to be speed limited and expensive as a WAN technology for SP backbones, plus the complexity of managing devices of multiple various technologies.

3. The most common layer 1 technology used with ATM is SONET/SDH (fiber interfaces) as illustrated by Maria.

4. The main factor that has driven ADSL to adopt the use of ATM was the ability to provision multiple PVCs on the same physical link (which can't be done via PPP alone), more over the invention of PPPoA and PPPoE solved the issue of requiring PPP (with all its addons) over such implementations.

BR,

Mohammed Mahmoud.

Manoj Wadhwa · ‎03-16-2008

Hi Maria/ Mohammed,

Really appreciate for all your time and patience in sharing such a userful infomation ... phew .. after reading this .. i stepped out for a break, had a glass of juice, refreshed myself and replying back :-) ... Indeed, i now agree with you i was asking things that are not at the starter level :-) ... but yes, i now have a handful bunch of information thanks to both of you who have been very gratious in sharing your knowledge and experience with me. I seem to have developed a grave interest to know more of ATM's but i guess i will be content with what i know for now and try to find ways if at all more expertise is needed for my current work. You guys are champs and make people who use this forum hope that someone is there to really help you out. Do remain that way always ... Thanks again and Cheers !!!

Regards,

Manoj

marikakis · ‎03-16-2008

Hello,

Yes, Mohammed is right about the SAR process. Generally, if you dig into such details you always find pros and cons in anything.

MPLS lookups where supposed to be faster than IP longest prefix match lookups. It is generally easier to do exact lookups (i.e. look for specific label) than trying to find the best route among others that matches destination. 20-bit MPLS lookups turn out (according to what I have read) not to be faster that LPM lookups in high performance routers (the chips that does those two I read is exactly the same, probably because of specific algorithmic techniques and hardware combinations (SRAM, DRAM, TCAMs) that have been discovered and deployed).

The ATM lookups are also exact lookups. The good thing with fixed cells is that it is simpler to provision buffers inside the core devices (that do not SAR) because you know all the units you switch are equal size and easier to have QoS with equal size segments (no big packets blocking small packets behind them). Buffer provisioning is not a minor issue. Buffers are in memory and memory technologies continue to be slow. The more memory operations, the slower the forwarding performance.

I could put another question. Ok, lets say that ATM is bad. Coverged backbones they want, they do not want one separate ATM network to be there, they do not want 2 separate networks for data and voice. Why do we need MPLS over Ethernet? I am not asking this in contrast to ATM, we forget about it (until we go back and borrow something from it). Pure IP over Ethernet has less overhead, LPM is not really slower than MPLS lookup, so why MPLS?

The main answer up to now is supposed to be new services. Maybe some security and isolation (until proven otherwise), maybe easier VPN provisioning (I LOL here, not many people are very comfortable with MPLS VPNs and the prerequisite knowledge, while many people will set you up a VC in no time, many will tell you they want nothing to do with troubleshooting customer routing between sites). Maybe traffic engineering, maybe a lot of things, until something else is found. I wait for the time in the future where bandwidth will be so plenty, that nobody will care about traffic engineering, QoS and the like (this sounds like the "I have a dream" :-). You will just drop your traffic in the network and it will reach destination in no time. :-)

Kind Regards,

M.

mohammedmahmoud · ‎03-16-2008

Maria,

Totally agree with you that MPLS enhanced speed over ordinary routing is just a myth these days, after having the modern routers forwarding plane done in ASICs. However in my experience in Service Providers environment if you have the adequate MPLS knowledge then VPN provisioning over MPLS is a lot scalable and easier than traditional layer 2 VPNs especially with large scale implementations. And i believe that AToM is one of the fun parts of MPLS, having whatever layer 2 technology you require over the same backbone, and last but not least all the other MPLS features (QoS, TE, ...). However i wish your dream may come true :)

BR,

Mohammed Mahmoud.

marikakis · ‎03-16-2008

Mohammed,

First of all, I cannot thank you enough for supporting my dream. LOL!

Yeah, my impression from discussions with colleagues is that people are fond of L2 MPLS VPNs (although I've read in some Cisco Press book that L3 VPNs have had great success). AToM can be a great success in the transitioning period towards the Ethernet dominating future I dream of. :-) As time goes by, it seems to me that the network converges to very few building blocks and alternatives. When the alternatives will have become less, then MPLS might have accomplished its mission. Yeah, this is pure science fiction. Just want to clarify that I am not against some technology. I believe everything is justified according to the time and reason it has been deployed. Networking has been and still is experimental. We learn from past mistakes or choices that were good at some time, but now do not help to solve current issues. And sometimes we go back. Like the thing you said about L2 MPLS VPNs. This seems to me as an enhanced way of going back to Frame Relay PVCs (no SAR there, label instead of DLCI). That's what I had in mind previously when I talked about going back and forth or in circles. You define the endpoints in AToM, don't you? Why is this easier than setting up Frame Relay PVCs?

What I have heard recently is a rumour about L1 VPNs with optics with the help of GMPLS. I read that this is like MPLS but only in the control plane (bye-bye MPLS header). I read somewhere else (in some Cisco Press book note) that a humorous person could refer to a leased line as an L1 VPN. And now I wonder, are we going back into an enhanced leased line? :-)

Anyway, it seems that we will be having lot of study and work for the future.

Kind Regards,

Maria