Sek, the first group, erlc type A followed that rule but i think the deadline was about 5 days for the 'resends' to reliable hosts. The new rules for the next set are 7 days to reliable.

-Uplinger

Assuming then the automation turns any resends into having deadlines of 2.8 days (40%). I noticed 5.6 days on the A type repairs (in a post someone snagging one I was waiting for).

You got it exactly right. Deduct the longest possible WCG job from the maximum 'reliable' return period and you have an approximate cache size maximum, bar poor TTC projections. So if wanting the A Types and wanting to continue running A types, and previously the device did them in 46 hours, the cache size limit becomes about 2 days. If on line and having reliable internet connection 0.3 to 1.0 days more than suffices.

Here is what we changed things to yesterday:

We are setting the deadline for the Type A workunits to 20 days. However, we are loading the workunits so that all copies are sent to reliable machines immediately (not waiting for only the resends to be sent to the reliable machines). This causes the 40% factor to take effect and the deadline for a specific result is reduced to 8 days (more if the server determines that the computer will need a little more time).

For the type B, we will set the priority so that they get loaded and sent before any type C workunits but they will still be sent to any machine. The deadline for these workunits will be 7 days. Re-sends will then be set to 40% of 7 days = 2.8 days.

Type C will be set to normal priority and have the standard 10 day deadline and re-sends will have the 4 day deadline.

Since A and B and C are having different deadlines I suspect there being 3 feeds that also caters for testing if a device is eligible for A or B or C.