Hello Ruwan, thanks for your additional explanation and see my comments inline.
> I think this has a considerable performance overhead, so better skip I/O > writes and reads as much as possible. Yes, this was also my idea. > chunk.size is the size of a chunk that will be written to the temporary > data (data source, which can contain in-memory data as well as disk > based data if the threshold exceeds) for example if the chunk.size is > set to 1024*64 then your buffer size is 64 Kb and the chunk.threshold is > the number of chunks that triggers a temp file, from there onwards. > Is this clear to you now? If you need further clarifications please let > me know... No, now everything is perfectly clear. Thanks! > BTW: Why do you need to keep the buffer size to 64 Kb, you may keep the > buffer size to 8 Kb or 16 Kb and vary the max chunk length for in-memory > store (chunk.threshold) as 8 or 4 respectively, and I don't think this > will affect performance a lot but will be better when dealing with small > messages. Otherwise you are assigning a 64 Kb buffer even for 1Kb or 2 > Kb messages. Yes, of course you are right. My example was just a fictive one to understand your parameters. > Default values for these parameters are 1024 and 8 respectively. Did you changed your default values? I think some of our message haven't been larger than 8kb but did get written to disc. I will experiment with this setting a bit. Maybe something like: chunk.size = 1024 chunk.threshold = 64 is what we actually want. Again thanks a lot for clarification. Regards, Eric _______________________________________________ Esb-java-dev mailing list [email protected] http://wso2.org/cgi-bin/mailman/listinfo/esb-java-dev
