Amy Krause EPCC Profiling OGSA-DAI Performance OGSA-DAI Technology Update GGF17, Tokyo (Japan)
11 May 2006http:// Outline Performance improvements from OGSA-DAI version 2.1 to 2.2 Comparing delivery of relational result sets encoded as WebRowSet (XML document) or CSV (comma- separated values) Delivery of binary data using SOAP Attachments Thanks to Bartosz Dobrzelecki and Ally Hume. The following diagrams are included in a paper which was submitted to the UK All Hands Meeting 2006.
11 May 2006http:// Delivery of SQL Results Measuring combined efficiency of ResultSet object serialization, transfer and deserialization. Results include both client and server times. © EPCC, The University of Edinburgh, May 2006
11 May 2006http:// Delivery of SQL Results Time spent in the server only, split into three phases: PreInvoke (Apache Axis parsing), Invoke (OGSA-DAI server work), and PostInvoke (message transfer to client). © EPCC, The University of Edinburgh, May 2006
11 May 2006http:// Delivery of Binary Data Time taken to transfer a binary file using Base64 encoded data inside the body of a SOAP message and as a SOAP attachment © EPCC, The University of Edinburgh, May 2006
11 May 2006http:// Delivery of Binary Data Time spent on server side split into phases, sending binary data inside a SOAP message and as a SOAP attachment. © EPCC, The University of Edinburgh, May 2006
11 May 2006http:// Delivery of SQL Results Execution time for scenarios fetching SQL results converted to XML and CSV data using two delivery mechanisms: delivery inside the body of a SOAP message and delivery as a SOAP attachment. © EPCC, The University of Edinburgh, May 2006
11 May 2006http:// Conclusions Version 2.1 -> Version 2.2 –Improved WebRowSet implementation, up to 35% performance improvement compared to v2.1 –New ResultSet to CSV activity, up to 65% improvement compared to WebRowSet v2.2. –New delivery via SOAP attachments activity Binary data transfer –Takes only 25% of the time to transfer a binary 8MB file using SOAP attachments –Improvements due to smaller data size and smaller SOAP messages –Limited by I/O performance rather than CPU