[ https://issues.jboss.org/browse/TEIID-2504?page=com.atlassian.jira.plugin... ] Steven Hawkins resolved TEIID-2504. ----------------------------------- Resolution: Done Most of the apparent performance gains stems from the prefetch, which needed changes to the way results are written/read - so that full result deserialization can be performed after the prefetch result is sent (a more simplistic approach, such was employed with MMX still would still largely put the deserialization cost in the request/response critical path). The changes also allow for a limited multi-batch fetch. Due to the abstractions surround our socket usage it is not easy to have the server reserve/unreserve memory with the asynch nature of the server to client writes. So a compromise is to allow approximately 3 (more if the actual memory footprint of the batches is smaller than the estimate) batches to be fetched together. This memory usage is essentially unaccounted for, so extreme amounts of concurrent response processing may require rethinking this approach - the most simplistic being to change the notion of the default fetch size, which is currently still 2048. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira