[FLASH-USERS] Big Problem??

[Tomasz Plewa]

Seyit -

Try to simplify your problem. If you run in parallel, can you confirm 
that the results differ when you run on a single processor?

If you run on many levels, how about a single block/level solution?

As a rule of thumb: try to simplify your problem. This may help you 
identifying the source of discrepancies. As you mentioned, there are 
many components involved, not only the application code.

Tomek
--
Seyit Hocuk wrote:
> Oh man,
> 
> I am using 3 different computers (with different architectures, 
> compilers, optimizations, environments). All three start differently 
> immediately from the beginning (difference in dt_hydro and refinement).
> Should I worry very very very much?
> I'm now doing a complete simulation and am going to check if my end 
> results also change.
> 
> I'm trembling by the thought.....
> 
> 
> 
> 
> 
> 
> 
> 
> 
> Paul M. Rich wrote:
>> Seyit Hocuk wrote:
>>> There is one huge problem that I noticed though. Normally I am always 
>>> doing simulations on my own computer or at the cluster available to 
>>> me. However, now that I am doing on this new computer with this new 
>>> compiler, my initial results are different and the difference seems 
>>> to grow bigger during the simulation. I use exactly! the same initial 
>>> conditions. But I immediately notice that refinement is different 
>>> (refines less quicker) and dtHydro is slightly different and growing 
>>> more different compared to my computer.
>>>
>>> Could this be due to the fact that the new machine is 64 bit and mine 
>>> is 32 bit and thus more precise? Although of-course the compiler is 
>>> also different as are the libraries (Hdf5 and Mpich) as is the 
>>> operating system.
>>
>> Yes, this most certainly can cause appreciable differences between 
>> runs.  Different computer architectures and compilers can and will 
>> result in different optimizations being used, and will pretty much 
>> ensure that the order of floating point operations will be changed.  
>> Depending on the problem, this can give very different-appearing 
>> results when output files are compared with sfocu, particularly if the 
>> problem has a lot of very sharp discontinuities or other small, 
>> rapidly changing, features.  If the initial differences are small, at 
>> about machine precision for double values, it is probably due to this.
>>
>> It also may be the case that one compiler was automatically 
>> initializing data differently, or that uninitialized data is being 
>> used in the problem at initialization time unintentionally.  If you're 
>> getting large errors at initialization, I'd check this.
>>
>> Hope this clarifies things for you,
>>
>> Paul Rich
> 
> 

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: tomek.vcf
Type: text/x-vcard
Size: 296 bytes
Desc: not available
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20080718/fabf9a56/attachment.vcf>