Last night 9/06/10, I received 6 rosetta WU's that ran
with priority (due date 9/16) so they "usurped" my
quad core machine for ~6 hours, displacing 4 other
running rosetta WU's with closer due dates (9/14),
as well as tasks for other BOINC projects that were
amicably sharing the CPU's according to my set
percentages. As soon as the priority WU's finished,
they disappeared from the machine - that, too was
highly unusual, because WU's usually remain resident
for minutes to hours before they report in. Those
WU's were in no danger of failing the report
deadline, so why run at priority?

I have very fast overclocked CPU's, and it looks like
somebody sitting in a lab at 9:00PM at night wanted
those 6 results right NOW, and dispatched them to me
for that reason, marked priority from the get-go.
I don't mind -
ordinarily I wouldn't be monitoring the machine so
late at night, but I'm using new hardware, a new OS,
and not-yet-stable overclocking, so I like everything
else to behave predictably. Anyone else have this
happen?
Stuart

---

This happened, because you had this long running modell. It was running double of the estimated time. The BOINC manager will apply that running time to all other Rosetta WUs. So your cache doubled (6 hours instead of 3 hours per WU). So you were probably facing some deadline problems and therefor the BOINC manager started running Rosetta tasks with high priority. A couple of WUs later the estimated running time dropped to 3 hours again and this solved the problem.

---

The project sets the deadline, the BOINC client determines which tasks to run and when to report results.

Because Rosetta was short on work last week, perhaps your machine built up some debt to Rosetta. Debt is where the CPU time spent on a given project falls short of the resource shares you have configured. This can cause BOINC to focus on a project that has been short on work for a few days after it comes back online. When your machine had no Rosetta work, it was doing more work for other projects. This is just the pay back of that "debt".

I believe you are describing the status shown in the BOINC Manager that it is running a task a high priority. Note that the actual run priority on your machine as compared to other non-BOINC applications is always LOW. This can happen if BOINC has some doubt as to whether the tasks will complete in time, which can be due to poor estimates on how long the tasks will take to run, or estimating on the recent ratio of time your machine is actually running BOINC (i.e. if BOINC has not been running recently for some reason, it begins to project in to the future that it doesn't run all the time, and hence will take longer to complete tasks). If you are tinkering with clock rates, BOINC is probably having trouble calibrating it's DCF and how long work will take to run. Adjusting the target work unit runtime (a Rosetta-specific option) could introduce such oddities as well.

As to why they ran before those with an earlier deadline... I can only say that they have been tweaking the methods used by the client to decide what to run and when, and I believe the current BOINC client code from Berkeley has resolved such scheduling issues.

In short, BakerLab did not control or orchestrate what you observed.

---

Rosetta Moderator: Mod.Sense