To introduce, the node in question runs ugm in about 5:30 hours, gets work which then lists this as 5:40 hours per assignment. The project current average is 3:35 hours i.e. it's a slow node. The puzzle is, time and again as in the below example, the amount of work sent divided by the number of tasks is a multiple of the real time. Here asked for 285k seconds, supposedly getting 297k seconds or 59.4k seconds per task, equating to 16.5 hours per assignment, yet the true estimated runtime per the agent is 20.4k seconds, not one third. This results in multiple work requests, multiple hits on the server until the cache has been topped up, but it's not! Running tasks plus 7 buffered is not adding up to 0.5 days per core setting. It's hardly 6 hours now, yet the remaining time total value is claimed to be 15:43 hours per core.

615 World Community Grid 11/7/2014 5:14:38 PM [sched_op] Starting scheduler request
616 World Community Grid 11/7/2014 5:14:38 PM Sending scheduler request: To fetch work.
617 World Community Grid 11/7/2014 5:14:38 PM Requesting new tasks for CPU
618 World Community Grid 11/7/2014 5:14:38 PM [sched_op] CPU work request: 285307.18 seconds; 0.00 devices
619 World Community Grid 11/7/2014 5:14:43 PM Scheduler request completed: got 5 new tasks
620 World Community Grid 11/7/2014 5:14:43 PM [sched_op] Server version 701
621 World Community Grid 11/7/2014 5:14:43 PM Project requested delay of 121 seconds
622 World Community Grid 11/7/2014 5:14:43 PM [sched_op] estimated total CPU task duration: 297056 seconds

629 World Community Grid 11/7/2014 5:16:48 PM [sched_op] Starting scheduler request
630 World Community Grid 11/7/2014 5:16:48 PM Sending scheduler request: To fetch work.
631 World Community Grid 11/7/2014 5:16:48 PM Requesting new tasks for CPU
632 World Community Grid 11/7/2014 5:16:48 PM [sched_op] CPU work request: 78592.81 seconds; 0.00 devices
633 World Community Grid 11/7/2014 5:16:52 PM Scheduler request completed: got 2 new tasks
634 World Community Grid 11/7/2014 5:16:52 PM [sched_op] Server version 701
635 World Community Grid 11/7/2014 5:16:52 PM Project requested delay of 121 seconds
636 World Community Grid 11/7/2014 5:16:52 PM [sched_op] estimated total CPU task duration: 118795 seconds

So why is it saying one when it's actually something quite different? The riddle may be in the value of <on_frac>0.458250</on_frac>. The device was on Linux for longer, then booted back to Windows. A confused agent to add even though computing 24/7.

Status now: 8 running ugm, each worth 5:30 hours, a buffer setting of 12 hours and 15:30 hours buffer value, each. No more buffered work since starting to collect thoughts on this gobbledygook, of course, just the 8 running ones are there, 7 finished while writing