Reducing the traffic bottleneck in cloud-based data management
Author
Fragkopoulos, Alexandros
Term
4. term
Publication year
2013
Submitted on
2013-06-06
Pages
56
Abstract
Mange virksomheder bruger i dag cloud computing for at få adgang til computerkraft og lagerplads uden selv at opbygge alt udstyr. Men de tøver ofte med en fuld flytning til skyen på grund af eksisterende, lokale ressourcer (som de ikke vil spilde), bekymringer om sikkerhed og tvivl om tilgængelighed. Dette projekt foreslår derfor en hybrid løsning, der kombinerer virksomhedens egne systemer med offentlige cloud-ressourcer. Løsningen bygger på cloud bursting (at flytte ekstra belastning til skyen ved spidsbelastning), men går et skridt videre med to elementer: en randomiseret routing-algoritme, der fordeler trafikken mellem flere nærliggende datacentre hos skyudbyderen for at undgå flaskehalse, og et lokalt Software-Defined Networking (SDN)-miljø, dvs. netværk der styres centralt via software, som vælger hvilke applikationer der midlertidigt skal flyttes til skyen ved spidsbelastninger. Når efterspørgslen stiger, vælger SDN’et de applikationer, der skal “burste” til skyen i en begrænset periode for at aflaste det interne netværk. Den randomiserede routing-algoritme vælger derefter tilfældigt et egnet datacenter fra en liste over datacentre i nærområdet, som skyudbyderen stiller til rådighed; overbelastede eller fejlramte datacentre kan udelukkes fra listen. Data replikeres kort før trafikken omdirigeres til skyen. Målet er at sprede trafikken, undgå flaskehalse (fx lager-I/O-flaskehalse, når for mange brugere rammer ét datacenter), og samtidig holde styr på udgifterne via betal-per-forbrug. Et konkret eksempel er en e-handelsløsning, hvor virksomheden som udgangspunkt kører på egne systemer med fuld kontrol over data og applikationer, men skalerer ud i skyen ved travle perioder. Rammeværket giver også skyudbydere mulighed for at indføre et ekstra båndbreddegebyr, hvis det valgte datacenter ligger tæt på virksomheden.
Many organizations use cloud computing to access computing and storage without building all the infrastructure themselves. Yet they often hesitate to move everything to the cloud because they already have on‑premises resources they do not want to waste, and they worry about security and availability. This project therefore proposes a hybrid approach that combines a company’s own systems with public cloud resources. It builds on cloud bursting (shifting extra load to the cloud during peak demand) and adds two ideas: a randomized routing algorithm that spreads traffic across multiple nearby cloud data centers to prevent bottlenecks, and an on‑premises Software‑Defined Networking (SDN) setup—network behavior controlled centrally by software—that decides which applications should temporarily move to the cloud during peak hours. When demand rises, the SDN selects applications to “burst” to the cloud for a limited time to relieve the internal network. The randomized routing algorithm then picks at random an eligible data center from a list of sites in the area provided by the cloud provider; overloaded or malfunctioning data centers can be excluded. The necessary data are replicated just before traffic is redirected. The aim is to distribute load, avoid bottlenecks (for example, storage I/O congestion when too many users hit one site), and control costs through pay‑per‑use. A use case is migrating an e‑commerce website: most of the time it runs on the company’s own infrastructure with full control over data and applications, and during busy periods it draws on cloud capacity as needed. The framework also allows cloud providers to apply an extra bandwidth fee when the chosen data center is geographically close to the enterprise.
[This abstract was generated with the help of AI]
Documents