Nobach, Leonhard (2018)
Seamless Flexibility in High-Performance Network Functions Virtualization.
Technische Universität Darmstadt
Ph.D. Thesis, Primary publication
|
Text
2018-11-05_Nobach_Leonhard.pdf - Published Version Copyright Information: CC BY-NC-ND 4.0 International - Creative Commons, Attribution NonCommercial, NoDerivs. Download (7MB) | Preview |
Item Type: | Ph.D. Thesis | ||||
---|---|---|---|---|---|
Type of entry: | Primary publication | ||||
Title: | Seamless Flexibility in High-Performance Network Functions Virtualization | ||||
Language: | German | ||||
Referees: | Steinmetz, Prof. Dr. Ralf ; Kellerer, Prof. Dr. Wolfgang | ||||
Date: | November 2018 | ||||
Place of Publication: | Darmstadt | ||||
Date of oral examination: | 26 October 2018 | ||||
Abstract: | Communication network carriers are challenged to continuously deliver higher-performance, more adaptable network services to even lower costs. Network Functions Virtualization (NFV) is an architectural concept aiming to decrease costs and increase flexibility of a network infrastructure. In an NFV architecture, network functions, which are traditionally executed on specialized appliance hardware, are executed on standard, inexpensive, and general-purpose servers. Furthermore, NFV applies cloud computing principles to the network functions implemented for standard hardware, enabling elasticity, flexibility and a fast time-to-market. For a sufficient flexibility, it is often desired that network function instances can be quickly moved between physical locations while they are in operation, which requires seamless state migration. Existing state migration mechanisms have been primarily designed for and tested in intra-datacenter situations. However, new concepts like carrier edge clouds and fog computing might require a state migration method for network function instances over long-distance links. The latters likely do not provide the throughput and latency available in a datacenter. We have identified that current methods can only migrate seamless in long-distance situations, if either the network function or the long-distance link is subject to low utilization. Furthermore, there are currently elasticity limits when using hardware acceleration for NFV environments. Due to the fixed set of commodity CPU and hardware acceleration resources on a computing node, either of the aforementioned resource types might become underutilized. Furthermore, the extraordinarily high performance of widely-available, inexpensive chipsets found in network switches could highly increase resource efficiency of network functions. However, the use cases of these chipsets are commonly limited in functionality, and it is unclear if a carrier-grade network function can be implemented by using them. In this thesis, we propose a seamless migration mechanism for virtualized network functions, which reduces the state migration traffic compared to the state of the art by omitting redundant information. Our evaluation shows that if compared to the state of the art, the reduction of the migration traffic allows an almost three-fold increase of the network function instance's or the link's utilization during migration, while completing the migration in only one third of the time. We propose an architecture which meets elasticity demands of network function implementations requiring heterogeneous processing resources like FPGAs, commodity CPUs, or in-network processing. We furthermore propose a method to quantify the benefits of elastic FPGA provisioning. Finally we investigate the functionality of a widely-used switching chipset in the context of carrier network functions, and conclude that all essential features of a Broadband Remote Access Server (BRAS) can be implemented using it. Overall, we show that we can improve flexibility through enabling NFV state migration over long-distance links, as well as resource efficiency via increased hardware acceleration utilization. |
||||
Alternative Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-81640 | ||||
Classification DDC: | 000 Generalities, computers, information > 004 Computer science 600 Technology, medicine, applied sciences > 620 Engineering and machine engineering |
||||
Divisions: | 18 Department of Electrical Engineering and Information Technology | ||||
Date Deposited: | 19 Nov 2018 11:00 | ||||
Last Modified: | 19 Nov 2018 11:00 | ||||
URI: | https://tuprints.ulb.tu-darmstadt.de/id/eprint/8164 | ||||
PPN: | 439051649 | ||||
Export: |
View Item |