I don't know how your system is setup and how you manage the queue.
My machines are not really fast - since they are not the youngest - however the fastest machine (i7 6700K) needs about 16 - 17 hours for each ARP1 WU; the other machines (Phenom II x6) need about 24 - 27 hours for each ARP1 WU; the Ryzen 7 2700 machines need between 18 and 22 hours for each ARP1 WU.
The current deadline is set to 7 days. If your machines cannot achieve an ARP1 WU within 120 hours (5 days), you should not select the ARP1 project for those machines.
In all cases, since there is no feeding trouble with ARP1, you shall not maintain a large buffer: < 2 days or better < 1.5 day.
Extending the deadline will cause troubles at WCG side, since the batches have to be kept available much longer and it takes a lot of storage resources.
Cheers,
Yves
---
PS: For each project, the minimal performance requirements are clearly defined and shall be taken into account by the members.
In all cases, if you cannot operate your machines 24/7, it is advisable not to compute ARP1 work, otherwise you will simply waste energy and computational time without really contributing. OPN1 and MCM1 are much more convenient for machines not running 24/7.

BOINC was intended to make use of spare CPU cycles when the device is on but (in CPU terms) idling. I hibernate my PCs at night when I'm not using them, and they will be often off during weekends and holidays. They typically work with no queue, though occasionally a short queue may form. If researchers want 24/7 availability they should buy their own supercomputer. If they want free computing power, then they need to work within the limitations of the BOINC model, and work efficiently. Increasing the time to complete WUs would reduce the number of unneeded resends. This is a source of inefficiency, since those hosts that are pointlessly working on resends due to the inadequate time allowance could be working on unprocessed WUs (a host's record will tell you if and when it is likely to return a result). It would also be a big help if the African Rainfall Project generated snapshots more often so that hours of work are not lost whenever a reboot is required.

Same here.
One of my computers I run like 6-9 hors per day and I just noticed the some of the workuntis have run out of time and some of them have being aborted by the project after have being crunched for 14 hours, what a waste.

I have lowered the Project Limits to 1 now and will keep an eye on it for now.

Yves

I also use appconfig.xml but without the fraction line. It works fine. What does that line do?

Mike

I wonder how much slow down. After trying out 100% ARP1 tasks, some CPU like AMD FX 4100 have almost no slow down, and some CPU like Ryzen 3900x are severely slowed down. Probably RAM bandwidth limited. Highly recommend 2 RAM sticks, equal size, for faster dual channel RAM on AM4, AM3, LGA1155, LGA1150.

Here are the per-task CPU hours, comparing a few ARP1 with other tasks mixed in, to all ARP1 tasks.

CPU          , RAM size & speed     , OS   , 1 to 3 ARP1, All ARP1
Ryzen 3900x  , 4x 8GB =32GB 2800    , Win10, 12.79 hours, 22.31 hours
Ryzen 2700x  , 4x16GB =64GB 2933 ECC, Win10, 16.20 hours, 25.56 hours
Intel i7-2600, 2x 8GB =16GB 1333    , Linux, 28.39 hours, 34.71 hours
AMD FX 4100  , 4x 8GB =32GB 1600 ECC, Linux, 26.03 hours, 26.51 hours