You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@felix.apache.org by "Richard S. Hall (JIRA)" <ji...@apache.org> on 2011/04/08 17:43:05 UTC

[jira] [Created] (FELIX-2909) [Framework] Resolver could be more efficient if it detected if fragments were present

[Framework] Resolver could be more efficient if it detected if fragments were present
-------------------------------------------------------------------------------------

                 Key: FELIX-2909
                 URL: https://issues.apache.org/jira/browse/FELIX-2909
             Project: Felix
          Issue Type: Improvement
          Components: Framework
    Affects Versions: framework-3.2.0
            Reporter: Richard S. Hall
            Assignee: Richard S. Hall
            Priority: Minor
             Fix For: framework-4.0.0


When fragment handling was moved into the resolver for the 3.2.0 release, it introduced a minor but not insignificant performance penalty. The old fragment approach was handled by the state and it was able to merge fragments into hosts at install time, which only had to be done once. The new resolver has to try to merge fragments each time there is a resolve operation. Further to correctly handle fragments, the resolver needs to build a more complicated candidate data structure. These two issues were the cause of the performance penalty.

Unfortunately, it is not possible to completely eliminate the performance penalty because these issues represent necessary processing. However, we could try to be a little smarter and only do the extra processing when fragments are actually present in the candidate graph. Related to this, they way fragments are greedily resolved in the new resolver also exacerbated this issue since it was always trying to resolve any existing fragments even if they weren't necessary for the current resolve. So, we can also be smarter here and only greedily resolve fragments if their hosts are being resolved.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

[jira] [Updated] (FELIX-2909) [Framework] Resolver could be more efficient if it detected if fragments were present

Posted by "Richard S. Hall (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/FELIX-2909?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Richard S. Hall updated FELIX-2909:
-----------------------------------

    Fix Version/s:     (was: framework-4.0.0)
                   framework-3.2.1

> [Framework] Resolver could be more efficient if it detected if fragments were present
> -------------------------------------------------------------------------------------
>
>                 Key: FELIX-2909
>                 URL: https://issues.apache.org/jira/browse/FELIX-2909
>             Project: Felix
>          Issue Type: Improvement
>          Components: Framework
>    Affects Versions: framework-3.2.0
>            Reporter: Richard S. Hall
>            Assignee: Richard S. Hall
>            Priority: Minor
>             Fix For: framework-3.2.1
>
>
> When fragment handling was moved into the resolver for the 3.2.0 release, it introduced a minor but not insignificant performance penalty. The old fragment approach was handled by the state and it was able to merge fragments into hosts at install time, which only had to be done once. The new resolver has to try to merge fragments each time there is a resolve operation. Further to correctly handle fragments, the resolver needs to build a more complicated candidate data structure. These two issues were the cause of the performance penalty.
> Unfortunately, it is not possible to completely eliminate the performance penalty because these issues represent necessary processing. However, we could try to be a little smarter and only do the extra processing when fragments are actually present in the candidate graph. Related to this, they way fragments are greedily resolved in the new resolver also exacerbated this issue since it was always trying to resolve any existing fragments even if they weren't necessary for the current resolve. So, we can also be smarter here and only greedily resolve fragments if their hosts are being resolved.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

[jira] [Closed] (FELIX-2909) [Framework] Resolver could be more efficient if it detected if fragments were present

Posted by "Richard S. Hall (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/FELIX-2909?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Richard S. Hall closed FELIX-2909.
----------------------------------

    Resolution: Fixed

I've applied a patch that does a better job of avoiding a penalty. Perhaps it could be further improved, but this is a good start.

> [Framework] Resolver could be more efficient if it detected if fragments were present
> -------------------------------------------------------------------------------------
>
>                 Key: FELIX-2909
>                 URL: https://issues.apache.org/jira/browse/FELIX-2909
>             Project: Felix
>          Issue Type: Improvement
>          Components: Framework
>    Affects Versions: framework-3.2.0
>            Reporter: Richard S. Hall
>            Assignee: Richard S. Hall
>            Priority: Minor
>             Fix For: framework-3.2.1
>
>
> When fragment handling was moved into the resolver for the 3.2.0 release, it introduced a minor but not insignificant performance penalty. The old fragment approach was handled by the state and it was able to merge fragments into hosts at install time, which only had to be done once. The new resolver has to try to merge fragments each time there is a resolve operation. Further to correctly handle fragments, the resolver needs to build a more complicated candidate data structure. These two issues were the cause of the performance penalty.
> Unfortunately, it is not possible to completely eliminate the performance penalty because these issues represent necessary processing. However, we could try to be a little smarter and only do the extra processing when fragments are actually present in the candidate graph. Related to this, they way fragments are greedily resolved in the new resolver also exacerbated this issue since it was always trying to resolve any existing fragments even if they weren't necessary for the current resolve. So, we can also be smarter here and only greedily resolve fragments if their hosts are being resolved.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira