Software-Defined Networking (SDN) is becoming the reference paradigm in both the industrial and academic worlds to provide the most advanced Traffic Engineering (TE) solutions for future networks. However, taking all TE decisions at the controller, in a centralized fashion, may require long delays to react to network changes. This may cause severe congestion issues, and countermeasures can become suboptimal or even obsolete in a highly changing dynamic environment. Distributed approaches may result in a good alternative to alleviate network congestion acting rapidly in face of rising congestion and network topology changes. With the most recent advancements in SDN programmability, some decisions can, and must, indeed be offloaded to switches. We present a model that permit to route elastic-demands in a general network topology adopting a hybrid framework to support a semi-distribute approach of the control plane, in order to deal with path congestion. This problem can be naturally modeled using Game Theory, where players are network nodes that want to send their traffic towards a destination by maximizing their throughput. However, the self-optimizing behavior of nodes leads to inefficiency and poor performances. To face this underutilization of the network resources, we can let the network controller to provide a guide to the users to a better social welfare. This scenario translates into a Stackelberg game where the SDN controller takes the role of a leader, which chooses a set of routing paths for the selfish users, that behave as followers. To overcome the complexity of the problem and meet the time requirements of a realistic setting, we propose heuristic procedures. We also discuss operations at switch side, showing how real-time congestion-based decisions to estimate the best path exploit the network resources better. We test our framework in a simulator for different network topologies and instance sizes. We show how our model achieves the desired balance between making global decisions and reacting rapidly to congestion events.
Il Software-Defined Networking (SDN) sta diventando il paradigma di riferimento sia nel mondo industriale che in quello accademico, in quanto in grado di fornire le più avanzate soluzioni di Traffic Engineering (TE) per le reti del futuro. Tuttavia, prendere tutte le decisioni di TE lato controllore, in modo centralizzato, può portare lunghi ritardi nella reazione ai cambiamenti della rete. Questo causa gravi problemi di congestione e, in un ambiente dinamico e altamente mutevole, le contromisure possono risultare non ottimali o addirittura obsolete. Gli approcci distribuiti rappresentano una buona alternativa per mitigare tali eventi, in quanto agiscono rapidamente di fronte a congestionamenti e possibili cambiamenti topologici della rete. Con i più recenti progressi nella programmabilità SDN, alcune decisioni possono e devono essere affidate agli switch. Presentiamo quindi un modello che permetta di instradare flussi elastici in una generica topologia di rete, adottando uno schema ibrido che supporti un approccio semi-distribuito del piano di controllo, al fine di fronteggiare la congestione sui percorsi. Tale problema può essere modellato utilizzando la teoria dei giochi, dove i giocatori sono rappresentati dai nodi della rete che intendono inviare il loro traffico verso una destinazione cercando di massimizzare il loro throughput. Il comportamento dei nodi mirato alla sola utilità personale si rivela inefficiente e conduce a scarse prestazioni. Per far fronte a questo sottoutilizzo delle risorse, possiamo far giocare l'amministratore di rete, al fine guidare gli utenti verso un migliore benessere sociale. Tale scenario può essere valutato in termini di un gioco Stackelberg in cui il controller SDN assume il ruolo di leader che sceglie un insieme di percorsi di routing per gli utenti egoisti, i quali si adeguano a tale scelta. Per superare la complessità del problema e soddisfare i requisiti temporali di un ambiente realistico, proponiamo delle procedure euristiche. Discuteremo anche riguardo le operazioni lato switch, mostrando come decisioni basate sulla congestione in tempo reale per la stima del percorso migliore sfruttino meglio le risorse di rete. Testeremo il nostro framework in un simulatore per diverse topologie di rete e di dimensioni di istanze, mostrando come il nostro modello raggiunga l'equilibrio tra prendere decisioni globali e reagire rapidamente a fenomeni di congestione.
Semi-distributed congestion-aware traffic engineering in software defined networks
Benedetto, Emmanuele
2020/2021
Abstract
Software-Defined Networking (SDN) is becoming the reference paradigm in both the industrial and academic worlds to provide the most advanced Traffic Engineering (TE) solutions for future networks. However, taking all TE decisions at the controller, in a centralized fashion, may require long delays to react to network changes. This may cause severe congestion issues, and countermeasures can become suboptimal or even obsolete in a highly changing dynamic environment. Distributed approaches may result in a good alternative to alleviate network congestion acting rapidly in face of rising congestion and network topology changes. With the most recent advancements in SDN programmability, some decisions can, and must, indeed be offloaded to switches. We present a model that permit to route elastic-demands in a general network topology adopting a hybrid framework to support a semi-distribute approach of the control plane, in order to deal with path congestion. This problem can be naturally modeled using Game Theory, where players are network nodes that want to send their traffic towards a destination by maximizing their throughput. However, the self-optimizing behavior of nodes leads to inefficiency and poor performances. To face this underutilization of the network resources, we can let the network controller to provide a guide to the users to a better social welfare. This scenario translates into a Stackelberg game where the SDN controller takes the role of a leader, which chooses a set of routing paths for the selfish users, that behave as followers. To overcome the complexity of the problem and meet the time requirements of a realistic setting, we propose heuristic procedures. We also discuss operations at switch side, showing how real-time congestion-based decisions to estimate the best path exploit the network resources better. We test our framework in a simulator for different network topologies and instance sizes. We show how our model achieves the desired balance between making global decisions and reacting rapidly to congestion events.File | Dimensione | Formato | |
---|---|---|---|
2021_9_Benedetto.pdf
non accessibile
Dimensione
2.62 MB
Formato
Adobe PDF
|
2.62 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/179909