An evaluation of compression and streaming techniques for efficient transfer of XML documents with Simple Object Access Protocol (SOAP)

Abstract

In SOAP, the entire XML object is generated on the server before it is returned to the client. This puts unnecessary strain on server systems in terms of both memory and CPU. The objectives are to find why SOAP does not allow streaming of responses, possible solutions to the problem and outline alternative transfer methods. Furthermore, compression techniques for a streaming SOAP environment are evaluated, as well as performance of streaming versus an alternative method of data retrieval. The feasibility study concluded that SOAP itself allows streaming of responses, but the HTTP binding does not. This binding specifies the issue of a HTTP fault code in case of a SOAP processing error, meaning the processing must be completed before a HTTP code can legally be issued. One alternative to streaming is using a Request/N-Response message pattern, and dividing the data over several responses. As HTTP only supports a Request/Response message pattern, implementing this is not possible. Either the HTTP binding must be rewritten to allow streaming of responses while processing a request or HTTP must be replaced with for example DIME as the transfer protocol for SOAP to overcome these problems. Tests are set up to find the most suitable compressor technique and to verify that streaming SOAP responses utilize server resources better then alternative transfer methods. Results show that bzip2 is the most suitable compressor technique. And that streaming utilizes memory considerably more efficient, especially with multiple clients connecting.