King: estimating latency between arbitrary internet end hosts
Proceedings of the 2nd ACM SIGCOMM Workshop on Internet measurment
Vivaldi: a decentralized network coordinate system
Proceedings of the 2004 conference on Applications, technologies, architectures, and protocols for computer communications
Stable and Accurate Network Coordinates
ICDCS '06 Proceedings of the 26th IEEE International Conference on Distributed Computing Systems
Understanding network delay changes caused by routing events
Proceedings of the 2007 ACM SIGMETRICS international conference on Measurement and modeling of computer systems
Towards network triangle inequality violation aware distributed systems
Proceedings of the 7th ACM SIGCOMM conference on Internet measurement
iPlane: an information plane for distributed services
OSDI '06 Proceedings of the 7th symposium on Operating systems design and implementation
Network-Aware Join Processing in Global-Scale Database Federations
ICDE '08 Proceedings of the 2008 IEEE 24th International Conference on Data Engineering
On the treeness of internet latency and bandwidth
Proceedings of the eleventh international joint conference on Measurement and modeling of computer systems
Hi-index | 0.00 |
The performance of large scale applications, such as those enabled by service-oriented, grid and cloud technologies, heavily relies on aspects related to the network topology and latency. As such, predicting the actual communication latencies is of high interest. The current state-of-the-art solutions to the problem of estimating the latency among distributed nodes comprise algorithms that tend to build upon the notion of network coordinates (NCs). Since network conditions change continuously, NCs need to be updated very frequently and, thus, are prone to oscillations. We present a variant of the pioneer NC algorithm, called Vivaldi, which encapsulates a change detection mechanism to prohibit NCs updates unless the network conditions change significantly. The contribution of this paper is twofold: first, to assess the impact of change detection and, second, to evaluate the NC algorithms in a realistic service-based environment where real measurements refer to large data transfers, contrary to current approaches that collect feedback from much smaller data transmissions, such as pings. The evaluation shows that our variant improves both performance and stability with less overhead.