Programmable, Expressive, and Agile Service Function Chaining for Edge Data Centers
Name: CRISTINA KLIPPEL DOMINICINI
Publication date: 23/08/2019
Advisor:
Name |
Role |
|---|---|
| MAGNOS MARTINELLO | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| MAGNOS MARTINELLO | Advisor * |
| VINICIUS FERNANDES SOARES MOTA | Internal Examiner * |
Summary: Abstract
The edge computing paradigm transfers processing power from large remote data centers (DCs) to distributed DCs at the edge of the network. This shift requires the ability to provide network functions virtualization (NFV) solutions that can efficiently manage and combine a large number of dynamic services in a resource-constrained DC, while ensuring that performance requirements are met. However, the routing mechanisms of traditional data center networks are not adequate for the dynamic composition of these services, because they are complex, rigid, subject to large delays in the propagation of control information, and limited by the size of switches routing tables. In addition, traditional service function chaining (SFC) solutions in the service overlay are often decoupled from routing decisions in the network underlay, and restrict path selection options by traffic engineering. In this way, the NFV orchestrator cannot explore the full capacity of the network.
To tackle these issues, this thesis investigates a programmable, expressive, scalable, and agile SFC proposal that allows dynamic and efficient orchestration of the network infrastructure of edge DCs. This proposal is composed of three interrelated solutions that exploit virtualization and programmability technologies of DC networks with commodity network equipment. The first one, called VirtPhy, is a programmable architecture that takes advantage of the topological properties of server-centric DCs for NFV orchestration. The second solution, called KeySFC, is a topology-independent SFC scheme that exploits the concept of fabric networks with a clear separation between: (i) edge switches based on software-defined networking (SDN) that classify flows for SFC; and (ii) core switches that implement a source routing mechanism based on the residue number system (RNS), which eliminates the need for routing tables. Finally, the third solution, called PolKA, is a mechanism that extends RNS-based source routing using Galois Fields of two elements, GF(2), so that the binary representation of the routing system is closer to the available operations in commodity network equipment.
As proof-of-concept, prototypes of all proposed solutions were implemented with production DC technologies, such as OpenFlow, OpenStack, Open vSwitch and P4. The results of functional and performance tests showed that the solutions can enable SFC in edge DCs in a programmable, expressive, scalable, agile and low cost manner. Thus, the proposed SFC scheme provides mechanisms to the NFV orchestrator that allow traffic engineering to make optimized decisions in the selection of network paths. This thesis also paves the way for exploring various RNS-based source routing properties in SFC schemes, which can provide features such as route authenticity, fast failure reaction, and forwarding without packet rewrite.
Keywords: network functions virtualization, software-defined networking, service function chaining, edge computing, data center networks, source routing, commodity equipment, residue number system.
