Tuesday, August 25, 2009

LTE Flat Architecture or Insane Architecture?

So every one is talking of how easy LTE is to work with all IP involved. Does making the whole network IP make LTE simpler than 3G networks? Now I am not an expert in 3G networks but I will still say LTE is no easy to deploy than finding a diamond in an ocean. And making it all IP doesnt make it simple too. Ok, so what do people mean by Flat? Is it IP everywhere which is making it flat? What about the protocols over IP? Lets try to analyze the below figure.

200908252031.jpg
All that nice lady wants to do is make some calls and read some emails on her phone. What we have is LTE Uu interface which is an air interface. UE communicates to eNB via RRC protocol. Above RRC we have something called UE Layer 3. This is where all the attach request/response etc are created. RRC is way to communicate to with eNB. Below it is the complex OFDM network. OFDM along with MIMO makes it more complex for my little brain. When the sweet lady switches on here mobile phone UE triggers some messages to get attached to the network. This is where the control plane communication begins. Packet goes to eNB. S1-AP is the protocol which eNB uses to communicate with MME. This S1-AP is over SCTP transport which is yet another protocol in itself.
MME communicates with SGW using GTPc protocol which is over UDP. So MME here needs to understand S1-AP over SCTP on one side and GTPc on the other side. More over it should be a powerful router to run normal routing and switching protocols (MPLS?). So far so good. Lets make it more dramatic. Bring on the PMIP over S5 interface. PMIP in itself is completely another technology. Till now we have protocols specific to LTE domain, but PMIP is generic protocol which is made to fit into LTE network. Thank god that PMIP is an IP over IP tunnel which makes it little simple. Mobile options sit over IPv6 which are put in UDP and transported over IPv4 between SGW and PGW (PMIPv6 over IPv4).
What happened till now? Just the control plane is established. Wait a minute. Where is the security? Charging Policy? Where are the registers? Well they are lying around some where and using "IP" for communication. Again, what happened till now? We have UE context established in eNB. eNB will map the air interface to particular tunnel in S1-U interface. MME has the UE context setup and has told SGW and eNB what information is required to run the user plane (tunnel id's, qos etc). Then we have user plane info exchanged between SGW and PGW which is PMIP based. So we have a mapping between S5 user plane (GRE key), S1-U (TEID's) and LTE Uu interface (God knows what).
Sweet lady wants to make a call. She dials in number and starts talking. User plane begins. Somehow data reaches eNB. eNB tunnels the data over GTP-U header to SGW. SGW removes the GTP header place GRE header and tunnels the data to PGW. PGW removes the GRE header and forwards the data to internet cloud. So simple! Right. I pity SGW here. The poor thing is struggling hard to decapsulate one header and encapsulate another.
Its not over yet. What about the routing protocols here? Many are suggesting the whole network to be based on MPLS cloud, which makes it more complicated. Even with all IP if somebody wants to understand LTE from end to end it will take days. This is one heck of network which looks more scary than a internet core network which is of BGP, OSPF or whatever greek and latin based.
Its just enough to understand one interface in LTE. Say if you want to master S11 interface which is GTP based reading GTP specs will just not get you there. We also need to understand what is happening before and after GTP to actually figure why GTP is behaving the way it is behaving. So I say download the whole 3GPP specs, make you own database and get insanely lost in the wireless network which also has too many wires :).
Oh wait wait. Dont go away, its still not over. Sweet lady is not done for the day. At this moment sweet lady opens her laptop and connects her phone to the laptop using USB. Now she is accessing her notebook over LTE network. Well USB has underlying RS232 techniques (? no idea) over which they make PPP to run. Now the packet first hops from the note book to phone over PPP and phone moves the packet further. Huh! Now it's done. Atleast I cant think any more. (3G/CDMA integration ?)
Tidbits : If you understand what I wrote above, it proves that direct tunneling cannot be established between eNB and PGW. It also proves the we just cannot integrate PGW and SGW into a single entity (When PMIP is used, else we should be able to get both in one device). S5 is the interface which SGW uses to communicate with PGW. S8 is the interface which visiting SGW uses to communicate with home PGW.
Thats it folks, I am still working on below post. Let me know how the above looks and feel very very free to correct me if I get things wrong.

14 comments:

Kostubh Sharma said...

Hi Santosh,
I would like to discuss on your statement "we just cannot integrate PGW and SGW into a single entity":

I think that we can still make it happen if we use GTP instead of PMIP over S5 interface.

Does my thinking correct?

- Kostubh Sharma

Santosh Dornal said...

Absolutely true! Only PMIP is the reason why these two entities are seperate, else they can be put in one device. Will correct the post. Thanks for pointer.

Sripada Kamesh said...

Hi Santosh,

I think its time for us to spend time on discussing about ipv6 also. As most operators are mandating their vendors to support ipv6, its a point to be considered.

Santosh Dornal said...

Yes, IPv6 is one thing to look at. But what is the need for having IPv6 in the core network? A UE may be get an IPv6 address but I dont see a need to completely have EPC on IPv6. Mandating IPv6 is one thing but implementation is another.

Look at the S5 interface above where PMIPv6 running. But my guess is vendor will go for PMIPv6 over IPv4 when it comes to deployment.

Sripada Kamesh said...

May be u are talking abt the near term effects. But in long run,operators have to concentrate on Pmipv6 over ipv6 only. But for now they can deploy with ipv6 within ipv4. Its not at all a permanent solution.

Coming to ur comment, r u saying there is no need of going for ipv6 implementation in EPC..???

Santosh Dornal said...

I will not say that IPv6 is not needed in EPC, but I will say that considering the pace at which LTE is deployed there will not be IPv6 any sooner in EPC.

The same applies to PMIP to. Initial deployments will be over IPv4 (trust me on this, my source is solid :-) ). Once we see everything is on IPv6 then we shall also see PMIP running on v6.

Sripada Kamesh said...

As far as i know verizon which is planning for near LTE deployment, is going for PMIPv6 over ipv6 between SGW and PGW. In this case, SGW sould act as MAG and PGW as LMA.

Santosh Dornal said...

That is basic idea. SGW=MAG and PGW=LMA. But IPv6 I am not sure and you can be correct. Unfortunately people dont give out these things so its quite tough to predict.

Kostubh said...

I am not sure about Verizon but most of the other companies are starting the deployment with PMIP over IPv4 and they surely have long term plans to move for PMIP over IPv6 finally.

Sripada Kamesh said...

Hi Kostubh,
Verizon has mandated that the UE should have dual stack. This makes the EPC vendors to go for PMIPv6, so that they can send ipv4 HOA or ipv6 HNP to UE. For the sake of flexibility they want their EPC should support ipv6 as well.

Tom said...

Hi Santosh,

If SGW and PGW are integrated, is the handset forced to get a new IP address on SGW handover? (as a PGW will handle a certain subnet)

Thanks for the great blog!

Santosh Dornal said...

Hi Tom!

Interesting question! My thought is that when SGW and PGW are intergrated the functionalities will also be integrated. So in this scenario if SGW has to change then obviously the PDN GW has to change which leads in a IP address change.

Ranjeeta Nath said...

In the above architecture diagram we have a view of all the protocols over each interfaces and how each is communicating.Its much informative and organised.It would be interesting if the protocol present in each element is also shown like MME has NAS and how interaction happens with NAS on MME and S-11 and to SGW.

Santosh Dornal said...

Working on it. A white paper should be out soon!