Skip to Main Content
 

Global Search Box

 
 
 
 

Files

ETD Abstract Container

Abstract Header

Designing High Performance and Scalable Distributed Datacenter Services over Modern Interconnects

Narravula, Sundeep

Abstract Details

2008, Doctor of Philosophy, Ohio State University, Computer Science and Engineering.

Modern interconnects like InfiniBand and 10 Gigabit Ethernet haveintroduced a range of novel features while delivering excellent performance. Due to their high performance to cost ratios, increasing number of datacenters are being deployed in clusters and cluster-of-cluster scenarios connected with these modern interconnects. However, the extent to which the current deployments manage to benefit from these interconnects is often far below the achievable levels.

In order to extract the possible benefits that the capabilities of modern interconnects can deliver, selective redesigning of performance critical components needs to be done. Such redesigning needs to take the characteristics of datacenter applications into account. Further, performance critical operations common to multiple datacenter applications like caching, resource management, etc. need to be identified and redesigned utilizing the features of modern interconnects. These operations can be designed and implemented as system services such that they can be provided in a consolidated platform for all other datacenter applications and services to utilize.

In this thesis we explore various techniques to leverage the advanced features of modern interconnects to design these distributed system services. We identify a set of services with high performance requirements that have wide utility in datacenters scenarios. In particular, we identify distributed lock management, global memory aggregation and large scale data-transfers as performance critical operations that need to be carefully redesigned in order to maximize the benefits using of modern interconnects. We further explore the use of these primitive operations and RDMA capabilities of modern interconnects to design highly efficient caching schemes which are critical to most datacenters. We present these components in a layered framework such that applications can leverage the benefits of these services through pre-defined interfaces. We present detailed experimental results demonstrating the benefits of our approaches.

Dhabaleswar, K Panda, PhD (Advisor)
P Sadayappan, PhD (Committee Member)
Feng Qin, PhD (Committee Member)
185 p.

Recommended Citations

Citations

  • Narravula, S. (2008). Designing High Performance and Scalable Distributed Datacenter Services over Modern Interconnects [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1218604342

    APA Style (7th edition)

  • Narravula, Sundeep. Designing High Performance and Scalable Distributed Datacenter Services over Modern Interconnects. 2008. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1218604342.

    MLA Style (8th edition)

  • Narravula, Sundeep. "Designing High Performance and Scalable Distributed Datacenter Services over Modern Interconnects." Doctoral dissertation, Ohio State University, 2008. http://rave.ohiolink.edu/etdc/view?acc_num=osu1218604342

    Chicago Manual of Style (17th edition)