From: David G. (JIRA) <ji...@co...> - 2012-01-20 18:12:13
|
[ https://jira.codehaus.org/browse/RVM-943?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=288954#comment-288954 ] David Grove commented on RVM-943: --------------------------------- Hi Jeremy. It looks like the attached patch is corrupted or in an unexpected format. Could you try again? thanks, --dave > MMTk HeapGrowthManager heap growth ratio computation has discontinuities > ------------------------------------------------------------------------ > > Key: RVM-943 > URL: https://jira.codehaus.org/browse/RVM-943 > Project: RVM > Issue Type: Bug > Components: MMTk > Environment: affects all. > Reporter: Jeremy Singer > Assignee: David Grove > Priority: Minor > Attachments: heapsize_patch.tar.gz > > > In HeapGrowthManager.computeHeapChangeRatio(), the current implementation determines the heap size change ratio by a lookup in the 2-dimensional function table (indexed by liveRatio and gcLoad). Given a current liveRatio X and gcLoad Y, the code finds the table rows and columns with nearest values above and below X and Y, then does interpolation from these table lookup values to determine the heap size change ratio. > However, there is a bug in the interpolation. If X (or Y, respectively) is exactly equal to a row (or column, respectively) label value, then the interpolation still happens, between values in rows (cols) either side of row X (col Y). This leads to discontinuities in the heap sizing function - see attached graphs. > The submitted patch suppresses interpolation (interpolation correction value becomes 0) in the case where X or Y falls on a label value exactly, so avoiding the discontinuity. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://jira.codehaus.org/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira |