Scheduling algorithms for web crawling

Abstract

This paper presents a comparative study of strategies for Web crawling. We show that a combination of breadth-first ordering with the largest sites first is a practical alternative since it is fast, simple to implement, and able to retrieve the best ranked pages at a rate that is closer to the optimal than other alternatives. Our study was performed on a large sample of the Chilean Web which was crawled by using simulators, so that all strategies were compared under the same conditions, and actual crawls to validate our conclusions. We also explored the effects of large scale parallelism in the page retrieval task and multiple-page requests in a single connection for effective amortization of latency times.