Is it possible to let the user decide, which applications he wants to crunch, like it´s possible at the Primegrid project? (www.primegrid.com)

Some people don´t have that amount of ram needed to crunch garli.

Greetings, Keep

Hi.
What are the needs for MARXAN, please?

Best regards,
Francois

____________
--
La voix de ma contrebasse: ZAC

Hi.
What are the needs for MARXAN, please?

Best regards,
Francois

We\'re not exactly sure yet. We\'ll post as soon as we know. The project using MARXAN will hopefully launch in January some time.

Thanks,
Adam

We\'re not exactly sure yet. We\'ll post as soon as we know. The project using MARXAN will hopefully launch in January some time.

GARLI
Disk - Download: 1M-1G Upload: 1M-1G
RAM - 300-400M
CPU - 30 minutes - many hours

Note: the progress bar for GARLI typically stops at 70%! In this case the workunit continues to crunch, however, so please do not be alarmed.

A few changes I'd like to see:

1. Separate the GARLI workunits into at least two different ranges of expected memory requirements. Let the participants disable getting any workunits from the ranges with the highest requirements, without also disabling getting any from lower ranges.

2. Check if it's possible for GARLI workunits running on computers with multiple CPU cores to send back information on the amount of memory BOINC is allowed to use, and the multiplier used for the estimated CPU time, and do this well before even trying to do much other computation, so that the server can use this information for deciding how many more workunits it can send that computer without forcing all of them into high priority mode, and therefore causing some BOINC versions to ignore the memory limits.

3. If any more work on GARLI-GPU workunits is planned, consider the idea of not even trying to use more GPU cores than it can find enough GPU memory for. Also consider the idea of limiting the speed of big bursts of accesses between the CPU part and the GPU memory, in order to avoid freezing any other CPU work during that time. Also read this thread on the new compilers Nvidia is planning to provide, although possibly only for the new 300 series of GPUs it is planning to start offering soon:

http://www.gpugrid.net/forum_thread.php?id=1406#13070

Note that if you manage to write a GPU application that only requires compute capability 1.1, you're likely to get quite a few new participants with Nvidia cards that only have that compute capability, since the other BOINC projects which can use such cards are rather limited; probably only one or two of them offering enough workunits to keep such cards busy. Such cards often have rather limited GPU memory, though, such as 512 MB.

4. Modify the section of GARLI that reports the progress percentage so that it always increases this percentage whenever GARLI writes a checkpoint. Then it should be easier for participants to see that the workunit hasn't stopped doing anything, even if the percentage of increase isn't very steady, and if big jumps are possible when GARLI decides not to do a second set of calculations.

5. Consider modifying GARLI so that when it gets too close to the deadline, it sends back the last checkpoint or two in addition to anything it has written to the normal output files so far, and ends execution. The server should then use those checkpoints to create new workunits that start from those checkpoints instead of the beginning of the original workunit.