|
" Vritti: "
Gandhi, Shruti
Viniotis, Ioannis
| Document Type
|
:
|
Latin Dissertation
|
| Language of Document
|
:
|
English
|
| Record Number
|
:
|
1106596
|
| Doc. No
|
:
|
TLpq2405541584
|
| Main Entry
|
:
|
Gandhi, Shruti
|
|
|
:
|
Viniotis, Ioannis
|
| Title & Author
|
:
|
Vritti:\ Gandhi, ShrutiViniotis, Ioannis
|
| College
|
:
|
North Carolina State University
|
| Date
|
:
|
2020
|
| student score
|
:
|
2020
|
| Degree
|
:
|
Ph.D.
|
| Page No
|
:
|
163
|
| Abstract
|
:
|
Inter-datacenter networks connect the geo-distributed datacenters of cloud provider(s), and many times also encompass the internet to provide connectivity. The nature of traffic on these networks is dependent on the cloud applications running in these datacenters. Hence the traffic is generally a mix of flows with and without deadlines, with known and unknown volume. The underlying wide area network (WAN) used to transfer these mix-flows is an expensive and congested resource. The providers today do not provide any guarantees to the traffic traversing the inter-datacenter WAN. In the literature, there has been little work done in guaranteeing deadlines to the traffic traversing the inter-datacenter WAN, while there has been no work done to provide tailor-made guarantees depending on the nature of traffic requirement while ensuring fairness among different ow types served in the inter-datacenter WAN environment. In this work, we propose the problem of identifying admission control, scheduling and routing decisions to provide deadline guarantees and fairness to mix-flows in an inter-datacenter WAN environment. We use linear programming (LP) to mathematically formulate the problem with the objective of maximizing utility that is a function of revenue generating ability of the ow types and fairness among all ow types. We propose a spatial-temporal traffic engineering systerm ‘Vritti', that can provide these guarantees. We propose and solve the static version of the problem where the future requests are known to the system in advance. We then propose a more realistic dynamic problem where the future requests are not known in advance. We propose four algorithms, namely, greedy, greedy-fair, selective-rescheduling and selective-rescheduling-fair to provide guarantees to mix-flows. We evaluate the effectiveness of these algorithms using extensive simulations. With selective-rescheduling, we achieve close to 69% acceptance rate of hard deadline requests at arrival rate of 10, and close to 100% acceptance rate for lower arrival rates. Using greedy-fair algorithm, we achieve close to 50% acceptance rate for hard deadline requests and close to 100% fraction of non-deadline requests allocated at arrival rate of 10. With selective-rescheduling-fair algorithm we strike a nice balance between the aggressive deadline traffic oriented approach taken by selective-rescheduling algorithm and fairness oriented approach taken by the greedy-fair algorithm. We also propose problem variations in the context of business models such as federated-cloud and multi-cloud. We solve the problems in these models by modifying greedy, greedy-fair, selective-rescheduling and selective-rescheduling-fair algorithms and evaluate the algorithms comprehensively using simulations.
|
| Subject
|
:
|
Computer engineering
|
| |