京都大学 / 新制・課程博士 / 博士(情報学) / 甲第23550号 / 情博第780号 / 新制||情||133(附属図書館) / 京都大学大学院情報学研究科通信情報システム専攻 / (主査)教授 大木 英司, 教授 原田 博司, 教授 湊 真一 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

Layer 4-7 network functions (NF), such as Firewall or NAPT, have traditionally been implemented in specialized hardware with little to no programmability and extensibility. The scientific community has focused on realizing this functionality in software running on commodity servers instead. Despite the many advancements over the years (e.g., network I/O accelerations), software-based NFs are still unable to guarantee some key service-level objectives (e.g., bounded latency) for the customer due to their reactive approach to workload changes. This thesis argues that Machine Learning techniques can be utilized to forecast how traffic patterns change over time. A network orchestrator can then use this information to allocate resources (network, compute, memory) in a timely fashion and more precisely. To this end, we have developed Mantis, a control plane network application which (i) monitors all forwarding devices (e.g., Firewalls) to generate performance-related metrics and (ii) applies predictors (moving average, autoregression, wavelets, etc.) to predict future values for these metrics. Choosing the appropriate forecasting technique for each traffic workload is a challenging task. This is why we developed several different predictors. Moreover, each predictor has several configuration parameters which can all be set by the administrator during runtime. In order to evaluate the predictive capabilities of Mantis, we set up a test-bed, consisting of the state-of-the-art network controller Metron [16], a NAPT NF realized in FastClick [6] and two hosts. While the source host was replaying real-world internet traces (provided by CAIDA [33]), our Mantis application was performing predictions in real time, using a rolling window for training. Visual inspection of the results indicates that all our predictors have good accuracy, excluding (i) the beginning of the trace where models are still being initialized and (ii) instances of abrupt change. Moreover, applying the discrete wavelet transform before we perform predictions can improve the accuracy further. / Nätverksfunktioner i lager 4-7 som t.ex. brandväggar eller NAPT har traditionellt implementeras på specialdesignad hårdvara med väldigt få programeringsegenskaper. Forskning inom datakomunikation har fokuserat på att istället möjliggöra dessa funktioner i mjukvara på standardhårdvara. Trots att många framsteg har gjorts inom området under de senaste åren (t.ex. nätverks I/O accelerering), kan inte mjukvarubaserade nätverksfunktioner garantera önskad tjänstenivå för kunderna (t.ex. begränsade latensvärden) p.g.a. det reaktiva tillvägagångsättet när arbetslasten ändras. Den här avhandlingen visar att med hjälp av maskininlärning så går det att förutse hur trafikflöden ändras över tid. Nätverksorkestrering kan sedan användas för att allokera resurser (bandbredd, beräkning, minne) i förväg samt mer precist. För detta ändamål har vi utvecklat Mantis, en nätverksapplikation i kontrolplanet som övervakar alla nätverksenheter för att generera prestandabaserade mätvärden och använder matematiska prediktorer (moving average, autoregression, wavelets, o.s.v.) för att förutse kommande ändringar i dessa värden. Det är en utmaning att välja rätt metod för att skapa prognosen för varje resurs. Därför har vi utvecklat flera olika prediktorer. Dessutom har varje prediktor flera konfigurationsvärden som kan ändras av administratören. För att utvärdera Mantis prognoser har vi satt upp ett testnätverk med en av marknadens ledande nätverkskontrollers, Metron [16], en NAPT nätverksfunktion implementerad med FastClick [6] och två testnoder. Den ena noden skickar data hämtad från verklig Internettrafik (erhållen från CAIDA [33]) samtidigt som vår applikation, Mantis, skapar prognoser i realtid. Manuell inspektion av resultaten tyder på att alla våra prediktorer har god precision, förutom början av en spårning då modellerna byggs upp eller vid abrupt ändring. Dessutom kan precisionen ökas ytterligare genom att använda diskret wavelet transformering av värdena innan prognosen görs.

O projeto de mecanismos flexíveis e eficientes para o posicionamento e encadeamento de funções virtualizadas de rede (VNFs) é essencial para o sucesso de Virtualização de Funções de Rede (Network Function Virtualization, NFV). A maioria das soluções existentes, no entanto, considera custos fixos (e imutáveis) de processamento de fluxos e de largura de banda ao posicionar as VNFs em Pontos de Presença da Rede (N-PoPs). Essa limitação torna-se crítica em redes NFV com fluxos cujos comportamentos são altamente dinâmicos e nas quais os requisitos de processamento e os recursos disponíveis nos NPoPs mudam constantemente. Para preencher essa lacuna, propõe-se o NFV-PEAR, uma plataforma para o posicionamento e encadeamento adaptativo de VNFs. O NFV-PEAR visa (re)organizar periodicamente os posicionamentos e encadeamentos de VNFs previamente determinados, objetivando-se manter um desempenho fim-a-fim aceitável mesmo durante flutuações nos custos de processamento e nos requisitos dos fluxos. Paralelamente, busca-se minimizar as mudanças na rede (por exemplo, a realocação de VNFs ou de fluxos) realizadas para cumprir esse objetivo. Os resultados obtidos, a partir de uma avaliação experimental, mostram que o NFV-PEAR tem potencial para reduzir significativamente o número de mudanças na rede necessárias para assegurar o desempenho fim-a-fim esperado para os fluxos, garantindo assim o funcionamento estável dos serviços. / The design of flexible and efficient mechanisms for proper placement and chaining of virtual network functions (VNFs) is key for the success of Network Function Virtualization (NFV). Most state-of-the-art solutions, however, consider fixed (and immutable) flow processing and bandwidth requirements when placing VNFs in the Network Points of Presence (N-PoPs). This limitation becomes critical in NFV-enabled networks having highly dynamic flow behavior, and in which flow processing requirements and available N-PoP resources change constantly. To bridge this gap, we present NFV-PEAR, a platform for adaptive VNF placement and chaining. In NFV-PEAR, network operators may periodically (re)arrange previously determined placement and chaining of VNFs, with the goal of maintaining acceptable end-to-end flow performance despite fluctuations of flow processing costs and requirements. In parallel, NFV-PEAR seeks to minimize network changes (e.g., reallocation of VNFs or network flows). The results obtained from an experimental evaluation provide evidence that NFV-PEAR has potential to deliver more stable operation of network services, while significantly reducing the number of network changes required to ensure end-to-end flow performance.

This thesis focuses on proposing a Software Defined Network (SDN) based programmable and capacity optimized backhaul and core network which is critical for 5G network design. Cloud Radio Access networks (CRAN) which is key enabler of 5G networks can address a number of challenges that mobile operators face while trying to support ever-growing end-users' needs towards 5th generation of mobile networks (5G). A novel layered and modular programmable CRAN architecture called Software Defined Virtualised Cloud Radio Access Network (SD-vCRAN) is introduced with Network Function Virtualization (NFV) and Software Defined Network (SDN) capabilities. The SDN-Base Band Unit (BBU) pool is shifted to the programmable core network site, where a centralised SDN controller manages the network servers and virtualised network function entities - Mobile Management Entity (MME), Serving/Packet Data Network Data plane (S/PGW-D), Serving/Packet Data Network Control plane (S/PGW-C), Software Network Defined Baseband Unit (SDN-BBU) and Local controllers (LC) via OpenFlow (OF) protocol. This approach simplifies network operations, improve traffic management, enable system-wide optimisation of Quality of Service (QoS) and network-aware application development. The control plane (excluding the preserved 3GPP standard interfaces: S1-MME, S6a, Gx) managed by the network servers provides load balancing, traffic management and optimisation tools for the data plane. The proposed work starts by reviewing the requirements of 5G networks, followed by discussion on 5G backhaul and core challenge. Then, an overview of CRAN, Evolved Programmable Core (EPC), SDN, NFV and related works. The simulation details of the proposed architecture are discussed along with the challenges faced by adopting SDN and NFV in mobile core. A thorough assessment of the interfaces and protocols that should be conserved or enhanced on both data and control plane is conducted. The result enables an architecture where the SDN-BBU pool shares a single cloud with the programmable EPC and the control plane is migrated from the network elements to a centralized controller, running on a virtual machine in the mobile core. The data and control plane separation removes overlaps and provides better signalling, as well as efficient network functioning to comply with latency demands. The proposed system performance is validated in terms of throughput, datagram loss, and packet delay variation under three scenarios: 1. single policy installation, 2. multiple policy installation and 3. load balancing. The load balancing performance of proposed system is validated comparing the performance of two different SDN controllers: Floodlight and OpenDaylight, where the later performs better in terms of throughput (no bandwidth restriction), packet loss (below 0.3%) and jitter (below 0.2ms). Furthermore, a detailed comparison of two SDN controller's - Floodlight and OpenDaylight performances is presented, which shows that OpenDaylight performs better only for less dense networks which needs less processing of messages without being blocked, and the Floodlight performs better in ultra-dense network. Some directions and preliminary thoughts for future work and necessary information to operators for building their roadmap to the upcoming technologies is presented.

Service Function Chaining (SFC) enriches the network functionalities to fulfill the increasing demand of value-added services. By leveraging SDN and NFV for SFC, it becomes possible to meet the demand fluctuation and construct a dynamic SFc. However, the integration of SDN with NFV requires packet header modifications, generates excessive network traffics, and induces additional I/O overheads for packet processing. These additional overheads result in a lower system performance, scalability, and agility. To improve the system performance, a co-optimized solution is proposed to implemented NF to achieve a better performance for software-based network functions. To improve the system scalability, a many-field packet classification is proposed to support a more complex ruleset. To improve the system agility, a network function-enabled switch is proposed to lower the network function content switching time. The experiment results show that the performance of a network function is improved by 8 times by leveraging GPU as a parallel computation platform. Moreover, the matching speed to steer network traffics with many-field ruleset is improved by 4 times with the proposed many-field packet classification algorithm. Finally, the proposed system is able to improve system bandwidth 5 times better compared the native solution and maintain the content switch time with the proposed SFC implementation using SDN and NFV.

O projeto de mecanismos flexíveis e eficientes para o posicionamento e encadeamento de funções virtualizadas de rede (VNFs) é essencial para o sucesso de Virtualização de Funções de Rede (Network Function Virtualization, NFV). A maioria das soluções existentes, no entanto, considera custos fixos (e imutáveis) de processamento de fluxos e de largura de banda ao posicionar as VNFs em Pontos de Presença da Rede (N-PoPs). Essa limitação torna-se crítica em redes NFV com fluxos cujos comportamentos são altamente dinâmicos e nas quais os requisitos de processamento e os recursos disponíveis nos NPoPs mudam constantemente. Para preencher essa lacuna, propõe-se o NFV-PEAR, uma plataforma para o posicionamento e encadeamento adaptativo de VNFs. O NFV-PEAR visa (re)organizar periodicamente os posicionamentos e encadeamentos de VNFs previamente determinados, objetivando-se manter um desempenho fim-a-fim aceitável mesmo durante flutuações nos custos de processamento e nos requisitos dos fluxos. Paralelamente, busca-se minimizar as mudanças na rede (por exemplo, a realocação de VNFs ou de fluxos) realizadas para cumprir esse objetivo. Os resultados obtidos, a partir de uma avaliação experimental, mostram que o NFV-PEAR tem potencial para reduzir significativamente o número de mudanças na rede necessárias para assegurar o desempenho fim-a-fim esperado para os fluxos, garantindo assim o funcionamento estável dos serviços. / The design of flexible and efficient mechanisms for proper placement and chaining of virtual network functions (VNFs) is key for the success of Network Function Virtualization (NFV). Most state-of-the-art solutions, however, consider fixed (and immutable) flow processing and bandwidth requirements when placing VNFs in the Network Points of Presence (N-PoPs). This limitation becomes critical in NFV-enabled networks having highly dynamic flow behavior, and in which flow processing requirements and available N-PoP resources change constantly. To bridge this gap, we present NFV-PEAR, a platform for adaptive VNF placement and chaining. In NFV-PEAR, network operators may periodically (re)arrange previously determined placement and chaining of VNFs, with the goal of maintaining acceptable end-to-end flow performance despite fluctuations of flow processing costs and requirements. In parallel, NFV-PEAR seeks to minimize network changes (e.g., reallocation of VNFs or network flows). The results obtained from an experimental evaluation provide evidence that NFV-PEAR has potential to deliver more stable operation of network services, while significantly reducing the number of network changes required to ensure end-to-end flow performance.

O projeto de mecanismos flexíveis e eficientes para o posicionamento e encadeamento de funções virtualizadas de rede (VNFs) é essencial para o sucesso de Virtualização de Funções de Rede (Network Function Virtualization, NFV). A maioria das soluções existentes, no entanto, considera custos fixos (e imutáveis) de processamento de fluxos e de largura de banda ao posicionar as VNFs em Pontos de Presença da Rede (N-PoPs). Essa limitação torna-se crítica em redes NFV com fluxos cujos comportamentos são altamente dinâmicos e nas quais os requisitos de processamento e os recursos disponíveis nos NPoPs mudam constantemente. Para preencher essa lacuna, propõe-se o NFV-PEAR, uma plataforma para o posicionamento e encadeamento adaptativo de VNFs. O NFV-PEAR visa (re)organizar periodicamente os posicionamentos e encadeamentos de VNFs previamente determinados, objetivando-se manter um desempenho fim-a-fim aceitável mesmo durante flutuações nos custos de processamento e nos requisitos dos fluxos. Paralelamente, busca-se minimizar as mudanças na rede (por exemplo, a realocação de VNFs ou de fluxos) realizadas para cumprir esse objetivo. Os resultados obtidos, a partir de uma avaliação experimental, mostram que o NFV-PEAR tem potencial para reduzir significativamente o número de mudanças na rede necessárias para assegurar o desempenho fim-a-fim esperado para os fluxos, garantindo assim o funcionamento estável dos serviços. / The design of flexible and efficient mechanisms for proper placement and chaining of virtual network functions (VNFs) is key for the success of Network Function Virtualization (NFV). Most state-of-the-art solutions, however, consider fixed (and immutable) flow processing and bandwidth requirements when placing VNFs in the Network Points of Presence (N-PoPs). This limitation becomes critical in NFV-enabled networks having highly dynamic flow behavior, and in which flow processing requirements and available N-PoP resources change constantly. To bridge this gap, we present NFV-PEAR, a platform for adaptive VNF placement and chaining. In NFV-PEAR, network operators may periodically (re)arrange previously determined placement and chaining of VNFs, with the goal of maintaining acceptable end-to-end flow performance despite fluctuations of flow processing costs and requirements. In parallel, NFV-PEAR seeks to minimize network changes (e.g., reallocation of VNFs or network flows). The results obtained from an experimental evaluation provide evidence that NFV-PEAR has potential to deliver more stable operation of network services, while significantly reducing the number of network changes required to ensure end-to-end flow performance.

京都大学 / 0048 / 新制・課程博士 / 博士(情報学) / 甲第22809号 / 情博第739号 / 新制||情||126(附属図書館) / 京都大学大学院情報学研究科通信情報システム専攻 / (主査)教授 大木 英司, 教授 守倉 正博, 教授 原田 博司 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

New services and ever increasing traffic volumes require the next generation of mobile networks, e.g. 5G, to be much more flexible and scalable. The primary enabler for its flexibility is transforming network functions from proprietary hardware to software using modern virtualization technologies, paving the way of virtual network functions (VNF). Such VNFs can then be flexibly deployed on cloud data centers while traffic is routed along a chain of VNFs through software-defined networks. However, such flexibility comes with a new challenge of allocating efficient computational resources to each VNF and optimally placing them on a cluster. In this thesis, we argue that, to achieve an autonomous and efficient performance optimization method, a solid understanding of the underlying system, service chains, and upcoming traffic is required. We, therefore, conducted a series of focused studies to address the scalability and performance issues in three stages. We first introduce an automated profiling and benchmarking framework, named NFV-Inspector to measure and collect system KPIs as well as extract various insights from the system. Then, we propose systematic methods and algorithms for performance modelling and resource recommendation of cloud native network functions and evaluate them on a real 5G testbed. Finally, we design and implement a bottom-up performance simulator named PerfSim to approximate the performance of service chains based on the nodes’ performance models and user-defined scenarios. / <p>Article 5 part of thesis as manuscript, now published.</p>

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第24941号 / 情博第852号 / 新制||情||143(附属図書館) / 京都大学大学院情報学研究科通信情報システム専攻 / (主査)教授 大木 英司, 教授 原田 博司, 教授 佐藤 高史 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM