Re: Fragmentation: Is there a solution yet?

Kunal Shah wrote:
> I found couple of postings about Bryan's intentions of creating an optional
> defragment switch or something similar as a solution to this. Can you tell
> me status on that.

Is this the post to which you refer?:
http://marc.theaimsgroup.com/?l=tini&m=99728218932372&w=2
or
http://lists.dalsemi.com/maillists/tini/2001-August/081281.html

Adding heap compaction capability would require rewriting the entire 
interpreter as well as most of the IO drivers.  I don't see this 
happening in the near future.

> Our application is doing bunch of file creation and deletion to save
> intermediate configurations/reports etc. And these I guess is leading to
> fragmentation problem causing insufficient heap. Also on file update we are
> also storing previous version of file so that tini can rollback to it if
> this update faces some problem. This will lead to have three copies of about
> 90 KB file at some moment which i guess would be a bad idea if memory cannot
> be defragmented. We can clear heap whenever it runs out of contiguous memory
> but that will cause it to download all the files again (using PPP) and lost
> all the configuration information too..We wanted to have that functionality
> used only extremely necessary (like heap corruption) which may not be that
> frequent. But our field devices have suffered this insufficient heap problem
> in just couple of weeks. So clearing out heap and downloading files again
> doesnt seem good solution here.
> We even haven't tried to do any remote updating on these devices yet.
> 
> I will greatly appreciate any hint/direction to solve this problem.

It sounds like you have 90K X 4 files (including your 88K executable) or 
about 360K of data in the filesystem.  Maybe the application could run 
System.gc() a couple of times after it jettisons some unneeded objects, 
then download/manipulate the new/old files.

-Bryan


If I only had some backing store and an MMU. :)

-- 
--
Bryan Armstrong
bryan.armstrong-6tN4nzCoH/[email protected]

_______________________________________________
TINI mailing list
TINI-6tN4nzCoH/[email protected]
To UNSUBSCRIBE, edit your profile, or see list archives:
http://lists.dalsemi.com/mailman/listinfo/tini