Here's a filter from BOINCview of the CEP1 version 6.19. There were 10 done, averaging 18:23 hours. The Quad averages 16 hours, the C2D near 23 hours. All Stock.

We are working on narrowing the range of how long the workunits last. Starting at batch 10 we have reduced the overall average length. We are hoping that these changes reduce it down to around 9 hours from the 15 hours the first batches.

You can identify the batch number from the workunit name by looking at the first part of the name. In this example:

E000003_352A_00002700y_1_3

'E000003' is the key part and means that this is part of batch 3.

However, there is larger variability of workunit duration then we have on our other projects so there is going to be more variation than normal. We will see if we can improve our estimates, but that may not be possible.

For the overall average estimate, our tool that estimates duration based on past results has enough data so it is now accurate. This should improve the estimates (subject to the variation).

With the length reduction, we have also reduced the deadline to 7 days. Due to the size of the returned results, we need to minimize the amount of time that they are stored on our servers before being sent back to Harvard. For a given batch of data, it takes about 1.5*deadline for the batch to finish. Somewhere around 80% of the workunits in a batch are returned in the first 3 days so those results have to be stored until it completes. We would like to reduce the size of the returned data, but Harvard currently needs all of it. We will continue to explore the issue with them but changes in this are not likely.

The Linux/Mac versions haven't launched yet. Perhaps that explains it?

Just to make sure, have you checked in the "Available Projects" section of your device profiles if "Please opt me in to new projects..." and/or "The Clean Energy Project" boxes are checked the way you think they are?

Cheers. Jean.

Many papers have been published on this subject.

It's not all that easy, though. Redundancy is very important in preventing data loss. Given the technical limitations of the BOINC system, it simply isn't practical.

In the case of World Community Grid, such a system isn't even necessarily desirable. The results need to be kept on the servers for validation purposes. After validation, the redundant data can be discarded. Finally, World Community Grid need to deliver the computed data to the researchers.

What the researchers do with the data is a different problem. Distributed storage using highly available desktops may be a very practical solution.