In my case at least the the time the computer is running per day was not considered. The WU went directly after download into high priority, still I wasn't able to return it in time, it wasn't sitting idle in my queue at all. Servers should have know that with all the information that is sent to them.
In particular from all that info in <time_stats>, the servers know better than myself, how many seconds per day this computer is running. The <on_frac> from my last request was 0.404740, that's around 9 hours and 40 minutes a day, so sending a WU, which needs 40+ hours with a 3.5 days deadline is probably not the best idea.


Anyway, the final state of this WU:

ARP1_0019494_060_4--  Valid	4/30/21 04:07:08   5/1/21 02:07:46    14.88   540.2 / 851.7
ARP1_0019494_060_2--  Valid 	4/26/21 16:04:09   4/30/21 16:00:42   30.12   640.8 / 851.7
ARP1_0019494_060_3--  Valid 	4/26/21 16:03:49   4/30/21 10:32:26   40.19   948.3 / 851.7
ARP1_0019494_060_0--  No Reply 	4/23/21 04:02:33   4/26/21 16:02:33   0.00    0.0 / 0.0
ARP1_0019494_060_1--  Valid 	4/23/21 04:02:31   4/26/21 20:53:16   29.05   755.2 / 851.7

In the same time those computers could have done two WUs, so maybe the current strategy to "speed up" patches, that are a bit behind the rest, needs to be revised.
Slightly less aggressive speed up with a 4.5-5.0 days deadline and _0, _1 and _2 would have been enough, except for _4, all other valid results have been returned just few hours after the 3.5 days deadline, so it's not just my old computer (if it was, I wouldn't post here, just turn off ARP).

I have a machine which need 4 days to crunch an ARP WU (2 checkpoints per day) and also receive 3.5 days units.

Maybe a rounding in the deadline : 3.5 rounded to 4 ? No margin ?

Ha, Ha, very funny. If you think "there is no solution", say "there is no problem".

It take 4 days for a 7 days deadline so it's not so "slow".

The trouble is only for a few WU.

I see two solutions :
- send it to the top 10% machines
- send it to the machine with 90 percentile which is under 3.5 days

Hi,

4 cores without hyper-threading.
ARP parts are already limited to 4 WU at the server side and no other project selected.
Hibernation is not possible.

So the cache is always empty even if it's set to 0.5 days and i download WU only if there is the time to accomplish the checkpoint before the shutdown.

No app_config on this machine. I dont need to limit concurrent WU.


It's always ok with the 7 days deadline because it need about 44 hours for each WU.
But with a 3.5 days deadline, i need sometime to run it more time than usual.

There is only one or two hours of margin so i can be considered as "reliable" but it's only because i micro-manage it.

bidon,
on a Phenom II x6, an ARP1 WU needs about 25-27 hours, and, like Mike, I limit ARP1 to 50% of the available threads. On an i7-6700K, it takes about 16-17 hours.
Cheers,
Yves