Πανεπιστήμιο Μακεδονίας
Orchestrating service function chains over resource-constrained network function virtualization (NFV) infrastructures
Abstract
dc:descriptionNetwork Function Virtualization (NFV) has recently emerged as a prominent cloud computing paradigm to address the scalability limitations of middleboxes, such as firewalls, load-balancers, and proxies. By decoupling them from their underlying hardware and fostering their implementation in the form of virtual network functions (VNFs), middleboxes can now be deployed on commodity servers in the form of virtual machines or containers and scale on-demand in response to evolving resource requirements. In addition, multiple VNFs can be organized into an ordered sequence to form a service function chain (SFC). Admittedly, VNFs and SFCs are the two primary constructs within the NFV ecosystem. While NFV renders middlebox deployments more elastic and consequently cost-effective, it also introduces several challenges, primarily related to the management and orchestration of VNFs and SFCs. A particularly significant challenge is the SFC embedding (SFCE) problem, which pertains to the optimal mapping of VNFs and virtual links to their physical counterparts, namely servers and links within a substrate network. The optimization of SFCE is NP-hard, with both exact and approximate methods proposed in the literature. However, as the NFV landscape evolves, leaning towards localized and resource-constrained compute models such as edge clouds, the relevance and scope of prevailing SFCE methods shall be revisited. To substantiate this, we identify four critical limitations of existing SFCE methods in resourceconstrained NFV. The first one stems from SFCE optimization over a single resource dimension, i.e., the CPU. While in core datacenters the CPU is indeed the main resource bottleneck, this might not hold in resource-constrained networks, given the high versatility of resource demands, along with the potentially lower capacity of remaining resource types, e.g., storage, memory, etc. Another limitation of prevailing SFCE has its grounds on its lack to handle imminent cross-service communications. The latter, which are particularly attractive in edge clouds, introduce additional constraints to the standard SFCE scope, and these necessitate substantial SFCE adjustments. A third limitation results from centralized reinforcement learning (RL) schemes that can be found in the respective SFCE literature. These typically make highly optimistic assumptions with regard to the observability of the underlying network, challenging their practicality in real-life deployments. To this end, we identify the need for the investigation of more appropriate RL schemes in the context of SFCE and NFV orchestrators in general. Last, not much light has been shed on SFC pre-processing methods. As such, SFCs are mostly perceived as invariable constructs, which contradicts their very virtual structure. In extension, SFCE solvers might struggle to obtain solutions on par with the optimization objective during unfavourable network states (e.g., high utilization). This raises the need for SFC pre-processing algorithms that can optimize the SFC structure according to fluctuating network conditions. This thesis sets out to explore the aforementioned challenges. Specifically, (i) we investigate SFCE across multiple resource dimensions as a means to minimize resource wastage in resource constrained NFV, proposing adept heuristic and exact algorithms to handle the intricacies of the problem at hand. Subsequently, (ii) we study SFCE through the lens of cross-service communications, which is a relatively new concept in the realm of NFV that advocates for crossservice interactions, thus amplifying network capabilities. Further, (iii) we examine multi-agent RL schemes within SFCE, paying particular attention to concise cross-agent communications and the interpretation of the scheme’s learning capacity across various network topologies. Last, (iv) we delve into SFC graph transformations in order to adapt the SFC request according to the resource conditions of the underlying network, thus enabling SFCE solvers sustain highquality solutions even in conditions of limited resource availability and/or high level of resource fragmentation.
Degree
thesis:*- Grantor dc:publisher
- Πανεπιστήμιο Μακεδονίας
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Pentelas, Angelos
- Πεντέλας, Άγγελος
Subjects
dc:subject × 14- Εικονικοποίηση δικτυακών λειτουργιών (NFV)
- Μαθηματική βελτιστοποίηση
- Ενισχυτική μάθηση
- Υπολογιστική παρυφής
- Network function virtualisation (NFV)
- Mathematical optimization
- Reinforcement learning
- Edge computing
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Μηχανική και συστήματα επικοινωνιών, Τηλεπικοινωνίες
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Communication engineering and systems, Telecommunications
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/54969