Kevin,

Quick question for clarification: Is the number of beta's per machine truly maxed at the number of cores per processor, or does it count virtual cores (hyper-threading) into that count?

Simply put, on an i7-920 which has 4 physical cores and can process 8 threads at once using hyper-threading, would the beta maximum be 4 or 8?

Thanks,

Hutch

System has no way to determine HT.
HT display 8 cores for OS.

Looks like the change worked well.

All three of my machines participating in the HCC beta. Each ended up with 1 or 2, including my slower linux box. From the discussion on the forum it looks like it had wide distribution.

It would be interesting to look at your Host/User numbers after the HCC Beta. The Host numbers should change. Hard to say if the user numbers will but I don't think it matters as long as more hosts are in the mix

And my quad is in that select group... out of left field [nowadays a Bushism] a ts02 came in of the ps family (A type). Put it ahead of the queue last night and been running 8 hours now and at 16% so it will finish 6 days before deadline... at about 47 hours run time. Just one of these babies equals 60+ C types. Mind you taking quick calc, for the project there are about 611 C types for every A type and much much easier to get, once the techs put 350V on the conveyor belt.

Hmm, <reliable_max_error_rate>0.001</reliable_max_error_rate> doesn't indicate the best possible error-rate is 0.001, even you're using a slide from the BOINC-2008-workshop, but this has negligible effects anyway so isn't worth discussing.

What kind of source you're using for handling errors in your nice, new table, is on the other hand still a mystery to me...

Now, according to your method to calculate, you'll example get, with 0.2% = 0.002 initial error-rate:
0.002 * 1.05 = 0.00201 and this is 0.201%, not a large difference, and one new validated task and you're reliable again.

My way to calculate on the other hand is this:
0.002 * 0.95 + 0.05 = 0.0519, this is 5.19%, and is a huge difference from 0.201%. Based on this calculation, you'll need 64 validated tasks to become reliable again.

As sources for my way of calculating error-rate, I've used the two links below:
This FAQ about Adaptive Replication also shows how the error-rate is calculated, and even it's using the more resent 0.1 and not the older 0.05 it should AFAIK be accurate in other regards.

Also, the code checked-into BOINC is here, there the old 0.05-code is shown in red, and the more resent code WCG possibly isn't using is 0.1 and green.

Oh well, since validation-errors is few and far between, in practice it likely doesn't matter how you calculate the errors...