Bug 27547 - Runtime workbench takes 15 minutes to come up
Summary: Runtime workbench takes 15 minutes to come up
Status: RESOLVED FIXED
Alias: None
Product: Platform
Classification: Eclipse Project
Component: Update (deprecated - use Eclipse>Equinox>p2) (show other bugs)
Version: 2.0.2   Edit
Hardware: PC Windows 2000
: P3 critical (vote)
Target Milestone: ---   Edit
Assignee: Christophe Elek CLA
QA Contact:
URL:
Whiteboard:
Keywords: performance
Depends on:
Blocks:
 
Reported: 2002-12-02 23:31 EST by Vadim Berestetsky CLA
Modified: 2002-12-16 07:40 EST (History)
3 users (show)

See Also:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Vadim Berestetsky CLA 2002-12-02 23:31:46 EST
In the latest WSAD drivers that are based on ECLIPSE 2.02 it takes 10 to 15 
minutes for debugger to come up when using  "Run-time workbench" launch config.

This was roughly 3 to 4 minutes 2 weeks ago.
Comment 1 Dejan Glozic CLA 2002-12-03 09:28:41 EST
Darin, this is 2.0.2. We didn't make any changes in launching code in 2.0.* 
stream. Ideas?
Comment 2 Darin Wright CLA 2002-12-03 09:33:30 EST
Are any breakpoints set? We know that method exit breakpoints are slow (a VM 
limitation).
Comment 3 Vadim Berestetsky CLA 2002-12-03 12:15:06 EST
It seems that initializing a significant number of plugins takes long time. 
WSAD has over 900 plugins.
Comment 4 Dejan Glozic CLA 2002-12-03 12:24:17 EST
I am curious if you are using any other PDE function prior to launching in 
debug mode. PDE must compute plugin paths and other information related to the 
target platform, so you normally pay a one-time price for discovering this 
information in the target plaform. For large products (with 900 plugins), this 
takes more than we would like to. However, as I said this is a once-per-session 
hit and should not happen every time you launch, and if you used function like 
plug-in import wizard, manifest editor, compute classpath etc., you already 
paid the price. After the process has been created, the rest of the startup 
time is due to the initialization of the second (target) instance.
Comment 5 Dejan Glozic CLA 2002-12-03 12:40:25 EST
We also noticed the problem and with the very recent integration builds. It 
only happens in debug mode and it is after the PDE launcher i.e. we launch the 
process fairly quickly, and return. The rest of the startup is in debug. Moving.
Comment 6 Darin Wright CLA 2002-12-03 12:48:52 EST
We will need the test case that causes the problem - i.e. is it WSAD or the SDK?
Comment 7 Vadim Berestetsky CLA 2002-12-03 12:52:10 EST
My feeling is that this has to do with a number of plugins. WSAD ships over 900 
plugins. I will try to limit the number of plugins and give it a try once I 
manage to download the latest WSAD driver :)
Comment 8 Lorne Parsons CLA 2002-12-06 08:59:41 EST
We ran into the same problem with WSADIE and have found where the problem is 
and a simple work around.

The problem is not with the number of plugins as previously mentioned in this 
bug report but with the number of Features. When starting the runtime-workbench 
in debug mode all of the time is spent reading/processing the feature 
information. We have monitored the files being accessed and paused the 
debugger, both pointed us at the features.

With this as our assumption, we renamed the existing features directory and 
then created a new features directory. After looking at the install.ini to 
determine which feature was the only one required, we copied this feature into 
the new features directory. The line in the install.ini 
is "feature.default.id=com.ibm.wsappdevie.win32", so the only feature in the 
directory is now com.ibm.wsappdevie.win32_5.0.0.

After making the above change, the debug mode startup time is 2-3 minutes which 
is acceptable based on the number of plugin currently in WSADIE (last time I 
checked it was 1130).

I hope this helps.
Comment 9 Darin Wright CLA 2002-12-12 15:21:28 EST
It sounds like we can say that this is not the "debugger's fault" then? If so, 
where should this bug be placed? It sounds like a performance issue for someone.
Comment 10 Jared Burns CLA 2002-12-13 22:35:00 EST
Reassigning to the Core. Hopefully, they'll know where this belongs.
Comment 11 Dejan Glozic CLA 2002-12-14 09:53:00 EST
Christophe, will our M4 performance improvements affect this problem?
Comment 12 Christophe Elek CLA 2002-12-16 07:40:43 EST
Action Taken: recent changes in M4 will help the issue
1) performance improvement when reconciling (bug 26947)
2) UI does not call UM anymore upon startup (bug 27358)
Action Plan: close