Experimental Evaluation of the PHP's cURL Library Performance

TL;DR

This paper experimentally evaluates the performance of PHP's cURL library, identifying it as the cause of download speed reduction from WebHDFS, rather than the API itself.

Contribution

It provides a systematic experimental analysis to isolate the performance bottleneck to the PHP cURL library, clarifying its impact on data transfer speeds.

Findings

cURL library causes download speed decrease

API performance remains stable regardless of file size

Experimental methods successfully isolate performance issues

Abstract

cURL (libcurl) is a popular and widely used library distributed with the php interpreter. It allows php applications to connect to and communicate with external resources (servers) by using wide variety of communication protocols. In most cases it is the preferred way of consuming external REST web services. Programmers usually use it for granted without even thinking of any performance issues. During an experimental analysis of the Hadoop's WebHDFS API throughput, it has been noted that read (download) speed from WebHDFS reduces with increasing the file size. However, this issue does not happen when writing to WebHDFS. Since the communication between the php application and the WebHDFS API is handled by the php's cURL library, then the cause of the download speed decrease could be either the cURL library itself or the API. This paper presents a series of experimental analyses aiming to determine the cause of the download speed decrease in previous experiments - whether it is the WebHDFS API or the php's cURL library. Both parties are tested in multiple ways separately and independently of each other. Results clearly prove (in two different ways) that the cause of the download speed decrease is the php's cURL library itself, not the consumed API.