This Bugzilla instance is a read-only archive of historic NetBeans bug reports. To report a bug in NetBeans please follow the project's instructions for reporting issues.
Summary: | [perf] Measure, compare and use mmaped files for data storage that currently occupies heap size | ||
---|---|---|---|
Product: | java | Reporter: | Jaroslav Tulach <jtulach> |
Component: | Unsupported | Assignee: | Jaroslav Tulach <jtulach> |
Status: | RESOLVED WONTFIX | ||
Severity: | blocker | CC: | issues |
Priority: | P3 | Keywords: | PERFORMANCE |
Version: | 4.x | ||
Hardware: | PC | ||
OS: | Linux | ||
Issue Type: | DEFECT | Exception Reporter: | |
Bug Depends on: | |||
Bug Blocks: | 44492 | ||
Attachments: |
Acode for reading current .cdx format and simulate potential mmaped trie format for different cell layouts
This patch replaces random access files by mmaps in FileCache. |
Description
Jaroslav Tulach
2004-10-07 12:20:23 UTC
Advice for now, very likely request for promotion e. IMO this should be either a task (given the summary), or a P3 defect. Since it is an "advice" for now, it should not need to be waived for 4.0. Note: By using mmap instead of RandomAccessFile, you also save a file descriptor (at least on linux), so you won't run out of them on large number of opened projects (like nbbuild/misc). Created attachment 18951 [details]
Acode for reading current .cdx format and simulate potential mmaped trie format for different cell layouts
Please remove the last three imports before using, I forgot to.... According to http://openide.netbeans.org/tutorial/reviews/opinions_44492.html this issue is TCR, definitively for promotion E. Created attachment 19168 [details]
This patch replaces random access files by mmaps in FileCache.
We have prototyped usage of java.nio in the b-tree and after measurements we decided to not use the java.nio there for now. We have also agreed to build an SPI for plugging different storage into the ClassIndex. We did it. Given this, for now - after agreement with Jarda - we are lowering the priority of this issue to P3. Jarda will try to implement the ClassIndex SPI using java.nio. Jardo? Do you plan to implement ClassIndex SPI? I still wish to. I cannot promise any date however. I have feeling that I am not going to work on this in near future. Thanks guys for all the measurements you did. If anyone anytime is about to try then please know that it is necessary to implement interface ClassIndexStorage and then provide own impl in constructor of ClassIndexu instead of "storage = new CISImpl(file);" put own "CISImpl". Reorganization of java component |