Tomohito Nakayama (JIRA) wrote:
This result suggests that problem around buffer at client is the reason why the patch does not improve streaming blob.
Is it correct to restate this as follows: - Before, both the server and the client had significant performance problems with the handling of large objects in the DRDA messages - After your patch, the server is better, but the improvement is hard to see because the client is now the bottleneck One way to see the overall improvement might be to measure overall system load during your benchmark, both before and after your patch. The expectation would be that overall system load drops due to the improved server-side efficiency, but throughput does not improve very much due to the remaining client-side bottleneck. If overall system load doesn't drop much at all, then would the conclusion be that the patch makes the server better, but the client has actually become worse due to using a less efficient algorithm to handle the no-length-header message format? thanks, bryan
