The popularity and proliferation of smart devices (e.g., smartphones, wearable devices, Internet-of-Things sensors) is resulting in an unprecedented growth in the amount of collected data. The current most popular approaches to manage this huge amount of data typically rely on cloud platforms located at the core of the infrastructure. As the number of devices and the amount of data they generate increases, such core-centric approaches are becoming increasingly inefficient as they need to transfer data back and forth between the core and the devices. Furthermore, the latencies associated with such data transfer are affected by the huge travel-distance needed to make the device communicate to the central cloud platform. To deal with the aforementioned situation new approaches have been introduced in both academia and industry, exploiting the power of the edge of the network to perform the computation closer to the data source. We noticed a discrepancy between the approaches proposed in research and in industry: research frequently assumes the possibility of running virtual machines or long-running containers on the edge. However, most real-world web infrastructure companies do not comply with this assumption due to the limited resource available in the edge. In this thesis we study the state of the art for stateful computations and data processing on the edge and after carefully analyzing the issues and the needs of the scenario we show the use cases predominantly affected by bandwidth and latency constraints. We then show the current frameworks available in the industry and notice how these solutions do not cover the use cases found. So we then propose a serverless approach effectively applicable by web infrastructure companies, that takes into consideration the problem of the scarcity of the resources, while still allowing quite powerful stateful computations on the edge. We also show how we implemented this new approach through a working prototype, and finally we investigate the gains developers may obtain by using our approach. We demonstrate how several use cases can benefit from this new system through discrete-event simulation, since running our prototype on an emulation of a global edge network was infeasible due to the sheer amount of resources needed to emulate even a small edge network.
La popolarità e la proliferazione di dispositivi intelligenti (e.g., smartphone, dispositivi indossabili, sensori Internet-of-Things) sta determinando una crescita senza precedenti della quantità di dati raccolti. Gli approcci attualmente più diffusi per gestire questa enorme quantità di dati si basano in genere su piattaforme cloud situate al centro dell'infrastruttura. Con l'aumento del numero di dispositivi e della quantità di dati generati, tali approcci basati su un core centrale stanno diventando sempre più inefficienti poiché devono trasferire i dati avanti e indietro tra il core e i dispositivi. Inoltre, le latenze associate a tale trasferimento di dati sono influenzate dall'enorme distanza di viaggio necessaria per far comunicare il dispositivo con la piattaforma cloud centrale. Per affrontare la situazione sono stati introdotti nuovi approcci sia nel mondo accademico che nell'industria, sfruttando la potenza dell'edge della rete per eseguire il calcolo più vicino alla fonte dei dati. Abbiamo notato una discrepanza tra gli approcci proposti nella ricerca e nell'industria: la ricerca presuppone spesso la possibilità di eseguire macchine virtuali o container di lunga durata sull'edge. Tuttavia, la maggior parte delle aziende di infrastruttura web non rispettano questa ipotesi a causa delle risorse limitate disponibili nell'edge. In questa tesi studiamo lo stato dell'arte per le computazioni con stato e per l'elaborazione dei dati sull'edge, e dopo aver analizzato attentamente le problematiche e le esigenze dello scenario mostriamo i casi d'uso prevalentemente affetti da vincoli di larghezza di banda e latenza. Mostriamo quindi i framework attuali disponibili nel settore e notiamo come queste soluzioni non coprono i casi d'uso trovati. Quindi proponiamo un approccio serverless effettivamente applicabile dalle aziende di infrastrutture web, che tenga conto del problema della scarsità delle risorse pur consentendo computazioni stateful abbastanza potenti sull'edge. Mostriamo anche come abbiamo implementato questo nuovo approccio attraverso un prototipo funzionante, e infine esaminiamo i benefici che gli sviluppatori possono ottenere usando il nostro approccio. Dimostriamo come diversi casi d'uso possono trarre vantaggio da questo nuovo sistema attraverso la simulazione a eventi discreti, poiché l'esecuzione del nostro prototipo su un'emulazione di una rete edge globale era impossibile a causa dell'enorme quantità di risorse necessarie per emulare anche una piccola rete edge.
Location-aware and stateful serverless computing on the edge
MOTTA, DENNIS
2020/2021
Abstract
The popularity and proliferation of smart devices (e.g., smartphones, wearable devices, Internet-of-Things sensors) is resulting in an unprecedented growth in the amount of collected data. The current most popular approaches to manage this huge amount of data typically rely on cloud platforms located at the core of the infrastructure. As the number of devices and the amount of data they generate increases, such core-centric approaches are becoming increasingly inefficient as they need to transfer data back and forth between the core and the devices. Furthermore, the latencies associated with such data transfer are affected by the huge travel-distance needed to make the device communicate to the central cloud platform. To deal with the aforementioned situation new approaches have been introduced in both academia and industry, exploiting the power of the edge of the network to perform the computation closer to the data source. We noticed a discrepancy between the approaches proposed in research and in industry: research frequently assumes the possibility of running virtual machines or long-running containers on the edge. However, most real-world web infrastructure companies do not comply with this assumption due to the limited resource available in the edge. In this thesis we study the state of the art for stateful computations and data processing on the edge and after carefully analyzing the issues and the needs of the scenario we show the use cases predominantly affected by bandwidth and latency constraints. We then show the current frameworks available in the industry and notice how these solutions do not cover the use cases found. So we then propose a serverless approach effectively applicable by web infrastructure companies, that takes into consideration the problem of the scarcity of the resources, while still allowing quite powerful stateful computations on the edge. We also show how we implemented this new approach through a working prototype, and finally we investigate the gains developers may obtain by using our approach. We demonstrate how several use cases can benefit from this new system through discrete-event simulation, since running our prototype on an emulation of a global edge network was infeasible due to the sheer amount of resources needed to emulate even a small edge network.| File | Dimensione | Formato | |
|---|---|---|---|
|
Executive_Summary_and_Thesis.pdf
accessibile in internet per tutti
Descrizione: Executive Summary and Thesis
Dimensione
2.82 MB
Formato
Adobe PDF
|
2.82 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/10589/182795