[FLASH-USERS] Fwd: info about FLASH

[Randy Hudson]

Some improvements coming for FLASH data in VisIt 2.0:


Begin forwarded message:

> From: Hank Childs <hankchilds at gmail.com>
> Date: March 21, 2010 10:07:46 AM CDT
> To: Randy Hudson <hudson at mcs.anl.gov>
> Subject: info about FLASH
>
> Hi Randy,
>
> I would appreciate it if you circulate information about VisIt
> improvements for FLASH data in V2.0.  V2.0 will be available in
> approximately two and a half weeks.
>
> Thanks!!
> -Hank
>
> There are four major changes:
> (1) Subsetting relationships were being represented inefficiently in
> VisIt and that data structure has been tuned significantly.  For big
> data sets, there used to be a large delay when creating plots, as well
> as a long pause after you pressed "Draw" and you started getting
> status (among other problems).  This is now signifcantly better.
> (2) The processor information is now available under database options.
> If you choose "File->Open" and set the file format to FLASH, you can
> then set the Default open options ("database options").  There is a
> check box there that allows you to access processor information (and
> not group information).  Putting the processor information in the
> default file format options was one of the changes necessary for
> achieving the speedup described in (1).
> (3) 1D curves were being resampled to the resolution of the finest
> patch.  This is now fixed and the curve is efficiently implemented.
> That said, there is a database option to use the old style curves in
> case there is a bug in the new implementation (AMR nesting is a tricky
> beast ... we think it works, but you never know).
> (4) The largest remaining problem in VisIt is with patch abutment.
> VisIt creates ghost data around patches at the same refinement level
> and it does this by using patch abutment information.  Its data
> structure (which is generic and not optimized for AMR) is taking
> almost 1GB in memory for FLASH data sets with hundreds of thousands of
> patches.  This data structure is needed for contours, but not for
> volume rendering or slices.  The database options now allow you to
> disable this structure, which should save you memory in the case where
> you have hundreds of thousands of patches.
>
> Best regards,
> VisIt development team


Randy.



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20100321/77d3a07b/attachment.htm>