New (Albert) application and workunits
24 Dec 2005 5:16:38 UTC
Topic 13199
(moderation:
I wanted to tell the dedicated crunchers a bit about the new application (called 'Albert') and workunits, that I have started testing on the public project today.
We've been doing private testing within the small group of Einstein@Home developers for several months, and are no longer finding problems and errors. So I have started to distribute a few thousand of these workunits to 'the general public'. If they work well we will start issuing primarily these workunits in the coming days.
A couple of key differences between the 'albert' (new) and 'einstein' (old) workunits.
- The new WU have different execution times, typically ranging from about 25% to 100% the previous execution times
- The new WU application incorporates all BOINC graphics and other bug fixes to date
- The new WU application has a slightly re-arranged screensaver, which includes our top wish-list item: a real time clock
I'll update (edit) this post if questions arise about how these new WU are structured. In many cases I'll then delete the post which asked the question, to keep the thread as compact as possible.
I have not forgotten that when we launched Einstein@Home in February 2005, we found a number of bugs because of the vigilance and sharp eyes of Einstein@Home users. So please call attention to strange behavior, either in this thread or in the Problems and Bug Reports message board.
Bruce Allen
[Edit Dec 24, questions from Paul Buck]
Quote:
1) We wouldn't have been lucky enough to get FLOPS counting this time around would we?
No, but I'll take a quick look at the API, and implement this if it's easy.
Quote:
2) Are the improved run times from optimized compiles for windows?
The compilation process is no more and no less optimized than before. The differences in run times come about because we are now using a sky search grid and frequency band which depends upon frequency. This makes it impossible for all workunits to be the same length.
Quote:
3) Is the Mac version still using Altiec?
Yes, the Mac version still uses Altivec optimization if the CPU supports the Altivec instruction set.
Quote:
I got one running right now by the way (thanks), and it is hard to tell over RealVNC, but the graphics looks like they are a little "prettier". Estimated run time is ~3 hours so that looks like about 25% of the prior (though I am only 13% through).
If you have a real-time clock in the upper right hand corner of the screensaver/graphics screen and the wording in the corners has slightly cleaner layout, then yes, you are running 'Albert'.
[EDIT 25 December, questions from various people]
Quote:
Will we be switching back and forth between Einstein and Albert apps?
Yes, for some time now, until we are sure that the Albert app is working as required.
Quote:
Does the Albert application have its own number (like 4.80) or is it still 4.79?
The Albert app has its own number and name. You will know you are running this by seeing what the name of the application is in the BOINC manager, or in the title bar of the graphics window. See the list of applications for more info.
Quote:
Is there any way we can download the new Albert application?
No. What work (and hence, what application) your computer gets is determined by chance. The 'scheduler' decides this when work is sent out.
[EDIT December 27]
Quote:
Is is intentional that the target number of results is three rather than the old value of four?
Yes, this is intentional. It may slow down result validation in some cases but will increase our computing power by ~ 25%.
RE: RE: For me it looks
)
Please, could you help us to identify which files are being modified? A simple way is to set your preferences to (say) 600 seconds, then monitor the timestamps of the files in projects/einstein.phys.uwm.edu/ and in slots/N/ to see which of these files is being written to more often than once every ten minutes.
The problem with the Linux
)
The problem with the Linux version of the albert application writing to disk more frequently than set by user preferences has been fixed. A new Linux version of the app (4.40) is now being distributed which fixes this.
For the technically-inclined, this was a nasty bug in the interface between the BOINC API library and the Einstein application. A API function called boinc_is_standalone() was returning a boolean value, but the API header file declared it to return an integer. It took a number of hours to track this down!
RE: I had an Albert WU that
)
Two other users have successfully completed the same WU, so there is probably nothing intrinsically wrong with it.