Alexandria Digital Research Library

Novel network architectures for cloud-based applications

Author:
Zhu, Yibo
Degree Grantor:
University of California, Santa Barbara. Computer Science
Degree Supervisor:
Haitao Zheng and Ben Y. Zhao
Place of Publication:
[Santa Barbara, Calif.]
Publisher:
University of California, Santa Barbara
Creation Date:
2016
Issued Date:
2016
Topics:
Computer science
Keywords:
Datacenter
Cloud computing
Wireless networks
Computer networks
Genres:
Online resources and Dissertations, Academic
Dissertation:
Ph.D.--University of California, Santa Barbara, 2016
Description:

Cloud-based applications demand significant, real-time computing power from the cloud. This means the underlying networks must provide high bandwidth and low latency data delivery. As the traffic demand grows exponentially, cloud-based applications have generated many challenges for network designers and operators, including 1) datacenter networks lack the performance to handle the increasing traffic demand, 2) datacenter networks are vulnerable to hardware failures and software bugs, and 3) mobile networks face bandwidth shortage.

Industries have spent significant efforts on expanding the networks by adding physical resources. For example, datacenter operators keep adding cables and switches, and mobile network operators purchase licensed wireless spectrum. However, this approach is facing fundamental limits, since simply adding more hardware leads to daunting monetary costs, and makes the network even less robust. Also, licensed wireless spectrum is limited.

Network architectures must evolve to provide dramatically better performance and robustness. In this thesis, we propose novel network architectures to address above challenges. One unique element of this work is the integration of new, emerging network hardware and technologies, like mmWave radios, RDMA-enabled NICs and programmable switches. For multiple scenarios, we successfully transform their distinct properties into drastic improvement over existing designs. Specifically,

1) To improve datacenter network performance, we propose to use flexible 60GHz links for addressing traffic hotspots, and end-host network stacks for high speed transmission.

2) We design a fault-isolated wireless control plane, and a data plane trouble shooting system, which captures and analyzes network traffic at packet level.

3) For wireless networks, we explore a dramatically different alternative in the form of 60GHz mmWave picocells with highly directional links.

These designs have promising outcomes. Inside datacenters, the novel infrastructure and protocols lead to significantly better throughput and latency, effectively reducing distributed job completion time by multiple times. Next, our datacenter control and data plane designs are able to maintain 99.99% connectivity even if 20% of hardware fails, and the recovering time is largely reduced. Finally, we achieve orders of magnitude larger wireless bandwidth and better sensing capability with 60GHz technology.

Physical Description:
1 online resource (252 pages)
Format:
Text
Collection(s):
UCSB electronic theses and dissertations
ARK:
ark:/48907/f3b56jw5
ISBN:
9781369339826
Catalog System Number:
990047190250203776
Rights:
Inc.icon only.dark In Copyright
Copyright Holder:
Yibo Zhu
Access: This item is restricted to on-campus access only. Please check our FAQs or contact UCSB Library staff if you need additional assistance.