Energy cost minimization and network management for data centers
Abstract/Contents
- Abstract
- Cloud-computing services are becoming integral part of our daily life. These services are supported by a type of computer system known as data center. As the demand for cloud-computing services soars, data centers are growing into massive-scale infrastructure of computational power, storage, and applications. Nowadays moderate-sized production data centers for enterprises typically comprise hundreds of thousands of servers, and consume tens of mega-watts of power to run and cool the equipment. Operating ever-growing data centers incurs soaring energy cost. Moreover, the sheer number of servers and networking devices in data centers presents a serious challenge to data-center network management. In this thesis, we are interested in problems on energy cost minimization and network management for data centers. The first part of the thesis concerns energy cost minimization for data centers. We present two workload scheduling methods for data-center energy cost minimization. The first is the temporal workload scheduling method for one data center. This method exploits the temporal variety of electricity price, and dynamically schedules user requests to execute in the time domain while guaranteeing a strict service delay bound for user requests. The second is the spatio-temporal workload scheduling method for distributed data centers in geographically separate regions. This method exploits both the spatial and temporal variety of electricity price, and dynamically schedules user requests to execute in both the space and time domain while guaranteeing a strict service delay bound for user requests. Evaluation experiments demonstrate that these two methods significantly reduce energy cost for one data center and distributed data centers respectively. The second part of the thesis concerns network management for data centers. An architecture named Ethane was recently proposed to manage enterprise networks, and data-center networks in particular. Ethane requires flow-based Ethane switches. One premise is that Ethane switches can be much simpler than conventional switches and routers. We build Ethane switches on the NetFPGA platform and install them as part of the Ethane-controlled prototype network deployed at Stanford University. Our experiments demonstrate that simple Ethane switches are sufficient to support the Ethane architecture for network management. In Ethane and other network management frameworks, statistics components are indispensable. An architecture named Counter Braids was recently proposed for network statistics. One premise of Counter Braids is that the hardware implementation of Counter Braids consumes little circuitry and supports fast counter-updating. We build a Counter Braids prototype on the NetFPGA platform and perform comprehensive tests. Our experiments demonstrate that Counter Braids uses a modest amount of resource; and the NetFPGA-based Counter Braids prototype easily gathers network statistics for an aggregate network throughput of 4 Gbps.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Publication date | 2013 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Luo, Jianying |
|---|---|
| Associated with | Stanford University, Department of Electrical Engineering. |
| Primary advisor | Prabhakar, Balaji, 1967- |
| Thesis advisor | Prabhakar, Balaji, 1967- |
| Thesis advisor | Mitra, Subhasish |
| Thesis advisor | Parulkar, Gurudatta M |
| Advisor | Mitra, Subhasish |
| Advisor | Parulkar, Gurudatta M |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Jianying Luo. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis (Ph.D.)--Stanford University, 2013. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2013 by Jianying Luo
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...