I see a lot of problems...

In my porting expiriance of several science applications for a lot of 
different projets, I saw a lot of differences in the results between the 
several CPU architectures.
In the worst cases, the science applications was complete unusable and 
produced complete invalid results. For example SETI produced on TRU64 
(Alpha), OpenBSD and NetBSD on SPARC  about 1000 times higher results. 
The problems was not fixed with different compilers nor any compiler 
flags. - I guess the root cause are problems in the math libs. Linux on 
Alpha, and Solaris or Linux on SPARC had not this problems.
Also some different compiler flags, could cause a lot of strange results.

Without any validation of the science application, is this improvment in 
my opinion complete unusable and unpractical.
 From my point of view, a opt-in third party or optimized app, which is 
tested and validated from the science projet, would been the better way. 
For example, the project test this app, put it on the BOINC server, and 
the user can select via the webpage the application they would like to use.

Lars

_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to