FERRET: Fall-back to LTE Microservices for Low Latency Data Access

Presenter: Muhammad Taqi Raza, Ph.D.The University of Arizona

June 25, 2020

1

Background - LTE

– LTE Evolved Packet Core (EPC) functionality is divided into control-plane and user-plane.

– Control plane logic performs device registration/deregistration, mobility, location update, paging, and many more.

– User plane forwards traffic to the next hop along the path to the selected destination network according to control plane logic.

2

Control-Plane

S1-AP MME HSS

Serving-Gateway

PDN-Gateway

User-Plane

LTE Core Network

Device Radio Network Internet

Background – IP Multimedia Subsystem (IMS) – Our Use Case Example

SGW PGW

LTE EPCRadio Network

IMS Core

Serving

Server

Media

Server

IMS Signaling (SIP)

IMS Service (RTP/RTCP)

Data Service

Telephony Network

Internet

IMS supports real-time multimedia services.

– Voice call request (control plane)

– Voice call speech packets flow (user plane)

– Regular user data traffic (user plane)

IMS Client

Proxy

Server

4

Existing IMS over LTE– Purpose-built hardware platforms, such as Ericsson’s Blade Systems (EBS), Alcatel-Lucent’s Element Management

System (EMS), etc.

– Coupling between software and hardware, such as Ericsson’s ERLANG, Alcatel-Lucent’s NVP, etc.

– Dedicated network resources for each user.

4

Device Radio Network

IMS

EPC

Internet

LTE-NFV Architecture

– LTE NFV virtualizes LTE core network functions over off-the-shelf boxes.– This reduces CAPEX and OPEX for operators.

5

Device Radio Network

Operator Network

vEPC Data Center

InternetRadio Network

Mang.

vEPCNetwork

Mang.

Why LTE-NFV Architecture is bad for IMS?

– Control plane signaling messages are executed on the cloud– Good: Different vNFs coordinate with each other to facilitate an event (e.g. mobility)

– User plane traffic is forwarded to Internet through vEPC Gateways (i.e. Serving Gateway, PDN Gateway, IMS NFs).– Bad: Voice traffic unnecessarily goes through the data center.

6

Device Radio Network

Operator Network

vEPC Data Center

InternetRadio Network

Mang.

vEPCNetwork

Mang.

Data Forwarding through Mobile Edge Compute NF

– Let’s decouple control-plane and user-plane– Control-plane traffic still goes to vEPC and vIMS– User-plane traffic is routed through mobile edge compute NFs.

7

Device Radio Network

Operator NetworkInternetRadio

Network Mang.

vEPCNetwork

Mang.

Mobile Edge Compute

– We decompose SGW and PGW user plane functionalities and install them at the edge– We also install MGW of IMS at the edge.

Design Considerations

– Decouple LTE and IMS CP and UP and install them at core and edge, respectively.

8

PGW – UResource management

IP address and TEID assignment

Packet forwarding

Forwarding of buffered packet

Accounting for Charging

Transport level packet marking

SGW – UResource management

IP address and TEID assignment

Packet forwarding

Transport level packet marking

UL/DL APN-AMBR enforcement

UL/DL bearer MBR enforcement

UL and DL service level gating

UL and DL service level MBR

PGW - C

IP address allocation

DHCPv4 / DHCPv6 client

Router advertisement

Mobility b/w 3GPP and non-3GPP

PGW pause of charging

Change of target GTP-U endpoint

SGW – C

Change of target GTP-U endpoint

Delay Downlink Data Notification

PGW pause of charging

Accounting for Charging

Bearer binding

Event reporting

Redirection

Mobile Edge

Forwarding of buffered packet

EPC and IMS CP

Proxy Server (of IMS)

Serving Server (of IMS)

Media Gateway (of IMS)

MEC Design

Design Considerations: Pure UP and CP Separation

– All UP functions are moved to edge, whereas all CP functions (including PCRF, LCS and others) are placed at the core.

Issue:– The UP traffic will steer to CP for charging. This adds latencies and bandwidth wastage for voice call operation.

9

Device Radio Network

Operator Network vEPC Data Center

Internet

Radio Network

Mang.

vEPCNetwork

Mang.

Charging

Provide copy of all common CP NFs at the edge

MEC Design

Design Considerations: Share Common Functions between CP and UP

– Those functions (e.g. PCRF or charging function) which both CP and UP rely should be shared (i.e. copied).

– Let the CP, being central entity, handle the NF allocation.Issue:– Creates race conditions. Example, SGW-U requests SGW-C for relocating SGW-U, while the TAU procedure is ongoing at the core, e.g.

MME performs SGW relocation by sending same request to SGW-C.

10

MMEDevice SGW-U

Relocation Req.(part of TAU)

SGW-C

Tracking Area Update Req.

Overload

Relocation Req.

Race Condition

Let UP manage its own resources based onpolicy provided by CP

MEC Design

Design Considerations: Share Common Functions between CP and UP

Those functions (e.g. PCRF or charging function) which both CP and UP rely should be shared (i.e. copied).

Issue:– Creates deadlocks. Example, when ModifyBearer request at CP and UP locks their respective PCRF copies.

11

Tracking Area Update

OverloadModify Bearer Req.

Session Modification

LockModify Bearer Req.

Session Modification

Lock

Device PGW-U PCRF-U PGW-C PCRF-CMME

Reflect update to PCRF-C

Reflect update to PCRF-U

Deadlock

Use blocking mode operations

MEC Design

Realizing IMS Application over MEC through FERRET

FERRET Key Idea

– Let the CP perform all control-plane operations at the core.

– Once all operations are performed at CP, then replay them at the edge.

– Only transmit VoLTE call control plane packets once the call is established (before user-plane traffic starts).

– VoLTE call requires:

(1) Call establishment phase (control plane), and

(2) Speech data packets flow (user plane).

Core

NF1 NF2M1

M2

NF2

NF1

M1 M2

End

Edge

IMS Application over MEC

14

Reducing User Plane Latencies through FERRET

13

– VoLTE call packets are forwarded to the Internet without going to the core.

– Baseline results represent operational VoLTE network latencies. – This includes eNodeB <-> vEPC <-> IMS latencies for both caller and callee

Summary

–MEC is part of 5G agenda that requires important network components to be installed at the edge.

– Through IMS, we demonstrate MEC architecture that reduces upto 6X user-plane latencies.

– In the future work, we will measure MEC design from different system and networking aspects.– Bandwidth, Latency, Call Rate, Mobility and more.

14