The recommendation stems frrom the early days of ARP. In those days most machines finished units in 15-27 hours.

If you shutdown a machine, any crunching after the most recent checkpoint is lost, which slows down the completion of the units.

Now, some machines can finish units in as little as 8 hours so would not have the same problems with shutdowns, although prolonged shutdowns for those units which have deadlines of 36 hours could still have problems.

3 or 6 day deadlines would be less of a problem except for the many slower machines still crunching here. Mine take 24 hours but crunch 24/7.

However, we have had problem units which have had their TimeStep shortened in order to overcome compatibility problems. Halving the TimeStep results in a doubling of the calculations required.

So, basically it is a matter of how fast your machine is, how long the shutdowns are and how long the deadlines are. The 24/7 recommendation is only a broad brush approach. See how you get on but be warned that deadlines might be exceeded if you shutdown.

The alternative is to use hibernate which retains the crunching after the last checkpoint.

Mike

I mostly use hibernate or sleep rather than shutting down completely and don't seem to lose tasks doing that. I do find that if when shutting down, I pause tasks and wait long enough to be sure they have stopped, then shut down the client before shutting the machine down it gets the fails down to under one in 50. I do the same if I ever have to shut down with CPDN which has had similar problems in the past though recent changes to the code for some models have greatly reduced the problem.

Many thanks for the report, Mike.

It is good to see the progress on the Extremes.

I am puzzled as to why there is no progress on the 3 ultra Extremes in generations 21 and 22. I assume that WCG hasn't released these units. I wonder why not?

Cheers,
Mark

There has never been any explanation for the 3 ultras.

Reducing the TimeStep has worked in other cases. Maybe they haven't tried a short enough TimeStep.

I suspect that they have restarted with the main extremes to start slowly and help them to catch up.

Mike

It looks like fewer ARP WUs are going out. You can see the decline in the number of results returned in the stats on the project itself as well as the status thread. The number in the image was up to 600s, but now it is in the 400s. I'm still getting a few ARP WUs here and there, so I don't see the decline personally, just in the data.