Well, that explains it. At least it is not at my end. Thanks for replying.

Ah cool . Just picked up a couple of Beta 6.10 WU's on two of my Linux boxes:
BETA_ CMD2_ 0001-1RKC_ A.clustersOccur-2IJM_ B.clustersOccur_ 12_ 1--
BETA_ CMD2_ 0001-1RKC_ A.clustersOccur-2DDF_ A.clustersOccur_ 15_ 2--
Hopefully all goes well.

Since this project has shown slight differences in results between Intel and AMD processors you might tell us which processor you are using. And possibly your operating system, Boinc version,...

Cheers. Jean.

So far I have gotten 6 Beta 6.10. Hopefully this will help give the techs a fair look at the AMD/Intel issue on the Linux platform as well.

3 WU"s (2 working, 1 waiting) AMD Athlon64 x2 RHEL5.3 BOINC 6..2.15
1 WU AMD Phenom 9600 (working) Fedora 10 BOINC 6.2.15
1 WU Intel Pentium D (working) Fedora 8 BOINC 6.2.15
1 WU Intel Centrino (working) Fedora 10 BOINC 6.2.15

Hello,
I'm the bioinformatician who actually built the "timeline" for each WU of the HCMD2 project. I'm not a WCG staff member, I'm part of the scientists team behind HCMD2.

For a wide range of projects, the computation time grows more or less linearly with some parameters. Thus, it's very easy to project the time taken by a given simulation. Unfortunately, this is absolutely not true with HCMD2 as molecular docking and Monte-Carlo strategies are involved.

After having computed some random steps of simulations, we elaborated our time projections for each couples of proteins from our dataset. Therefore, the first thing to understand, is that this projection is build upon a very very very tiny part of the whole simulation (if we were able to compute the whole thing, why would we need WCG and its generous volunteers ?). The second thing to understand is that the total number of steps needed to achieve our simulation cannot be predicted without actually doing the whole simulation (well, this won't be a *prediction* anymore, isn't it ?). Some part of the simulation take a long moment and some other parts compute faster. Therefore, if the time projection was done using data from a rather "quick part", the actual whole computation will be much longer than predicted. Conversely, a time projection done using data from a "slow part" would be overestimated.
At last, given the total time projected for a protein couple, we projected in how many WU the simulation must be split in order to complete each WU in "a reasonable amount of time" for current CPUs. According what I told you about the total time estimations, you can easily figure out how a "predicted reasonable amount of time" can become an "actual unreasonable amount of time".

I hope it helped to clarify why there are such time variations between WU and why, for some WU, you can't rely on the elapsed time to estimate the remaining time given the percentage of completion of the WU.

--
Michaël HEYMANN