Sunday, November 8, 2009

Evaluating LTE SGW

We know that Serving Gateway is one of the most important entities in LTE network. Say if XYZ service provider is evaluating a SGW, what are the factors that XYZ would be looking at. Lets avoid price and technical support offered by the equipment vendor here.  

Serving Gateway is supposed to handle both control plane and user plane traffic. Just to increase the complexity, lets consider that S5 interface is PMIP based. So we have GTPv2 communication on S11 interface, GTP-U encapsulated traffic on S1-U interface, PMIP based control plane and GRE encapsulated user plane traffic on S5 interface. So the technical requirements are revolving around control and user plane.

Control Plane: Input to the SGW are GTPv2 messages and output are PMIP message towards PGW. In between SGW needs to communicate with PCRF/Radius servers for pulling other information. So first thing XYZ wants to see is a successful session established. That means we need to get the basic functionality correct. More over XYZ would also want to see how easy it is to configure the SGW. This may not be one of the criteria for evaluating the box but its good have a easy to use interface. Getting the basic functionality right means there needs to be a successful interop. That is the SGW should work with other vendors MME or eNB or PGW or etc. This also means that SGW vendor should get the specs right :-).

Next, XYZ wants to see how many sessions can be activated per second. Also XYZ would be very interested to see how many active sessions can the SGW handle. May be a million users per box? That means SGW should have million contexts activated and data base integrity should be smooth. XYZ might also want to see how many dedicated bearers can be activated for single default bearer. More over XYZ might fancy to see how many default/dedicated bearers can be activated and deactivated per second. This test might prove how strong is the box is.

Handovers is another functionality which XYZ wants to see. Does SGW support various handovers mentioned in the spec. If so XYZ would prefer a number of users that can have successful handover per second.

Roaming too is important. There is typical requirement which arise with roaming scenarios. MME of visiting network might want to contact home network SGW for authentication, charging policy etc. Which means the GTP message is hopping for one router to other in the internet. Which also means that there needs to be some security involved. IPSec comes into picture. MME can encrypt the GTPv2 control messages in IPSec and pass it on to the home network SGW. This means SGW should also be to handle several IPSec sessions. I dont want to go below this layer. SGW might also be able to do BGP/OSPF/MPLS etc at the network layer. These could be some of the control plane requirements.

User Plane: First thing that comes into my mind is throughput. XYZ wants to see how may user plane session can be handled at what throughput. Usually the packet size is fixed in S1_U interface, so there is need to know the maximum throughput that box can achieve. I strongly believe that this is the most critical factors.

Finally, XYZ wants combination tests, thousands of sessions established along with thousands of user plane tunnels with different kinds of traffic. These are system level tests. If everything matches XYZ's requirement then they are ready to spend heck loads of money on the gateway.

These were few thoughts running through my mind. Feel free to add more!


Vikram said...

Dear Santosh,

I Bangalore based developer, this my LTE blog

Santosh said...

Thats cool vikram! I shall follow it.

Nickname unavailable said...

I believe SGW doesn't speak with PCRF. That is a PGW function via Gx(aka PCEF) for subscriber rate-limiting, route manipulation, content-based billing, etc. SGW values would be propogated to the SGW by the MME (aka softswitch) and would be propogated via some values specified wtihin the HSS/HLR.
I do believe there are some cases where the PCRF signals a bearer modification towards the network as well. However, it would have to be signaled via S5 and propogated forward from there.

Santosh said...

Thanks for info. I was guessing that SGW would talk to PCRF in case of PMIP based S5/8 interface rather than PGW. Right?

Yes, PCRF signals a bearer modification, which can result in creation of dedicated bearer or modification of a bearer. However I still dont understand how and what will make PCRF trigger a bearer modification.

Vijay Nag said...

exactly santosh. The interface is called gxx when PMIP is the interface between SGW and PGW

Anonymous said...

The PCRF can signal a bearer modification based upon any logic it is aware of like subscriber volume counters, matched policies based upon protocols like peer-to-peer, etc. Some PCRF vendors may exploit an API which allows users self service to upgrade provisioned AMBR (aka speed tiers).