DMCA
Efficient Numerical Error Bounding for Replicated Network Services (2000)
Cached
Download Links
| Venue: | In Int. Conf. on Very Large Databases (VLDB |
| Citations: | 42 - 5 self |
BibTeX
@INPROCEEDINGS{Yu00efficientnumerical,
author = {Haifeng Yu and Amin Vahdat},
title = {Efficient Numerical Error Bounding for Replicated Network Services},
booktitle = {In Int. Conf. on Very Large Databases (VLDB},
year = {2000},
pages = {123--133}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
The goal of this work is to support replicated network services that accept updates to numerical records from multiple wide-area locations. Given the high overhead of maintaining strong consistency, many replicated services can tolerate divergence of their shared data, as long as the numerical error is bounded. Target distributed services include replicated stock quotes services, online auctions, distributed sensor systems, wide-area resource accounting and load balancing for replicated servers. We present two algorithms to efficiently bound absolute error using only local information. Split-Weight AE separately bounds increases and decreases, while Compound-Weight AE bounds them together. The two algorithms can be combined to provide good performance and low space overhead. Our Inductive RE algorithm transforms relative error to absolute error solely based on local knowledge, taking advantage of the fact that the divergence was properly bounded prior to each invocation of...
Keyphrases
efficient numerical error bounding replicated network service network service numerical record replicated stock quote service replicated server local information absolute error local knowledge load balancing good performance split-weight ae compound-weight ae multiple wide-area location online auction wide-area resource accounting sensor system numerical error high overhead low space overhead inductive re algorithm transforms strong consistency
