Default is only ONE (1) day. btw

Have you used a BOINC 7.x.x version on a trickle server yet? You need to increase your "Connect about every" setting or no work will download until your cache runs dry. POEM work is currently available only 1 or occasionally 2 tasks per work request and frequently none, so with report results immediately, BOINC will request new POEM work every time it finishes a POEM task. Otherwise GPU will run out of work very quickly. Bit of a workaround but bigger caches will become more common with the work fetch policy of 7.x.x versions, particularly for those with unreliable internet connections.

That's strange, you have a silver badge for beta here. Anyway you'll find out how the new work fetch policy works for yourself if it is included in a release version.

I meant beta is not a production environment. People were asking about how BOINC 7.x.x worked with projects other than POEM GPU and I replied with my experience as did others. BOINC 7.x.x is required to process POEM OpenCL tasks, BOINC 7.0.11 is required to identify HD 7970.

OT, but the main problem with 7.0.xx and WCG only crunchers is that WCG only gives 15 tasks max per call and the Connect Every xx days [CE] & Additional Buffer [AB] have assumed a role of minimum / maximum buffer. This means that any work call that lifts the cache above the CE value, ends work fetching. With WCG defaults and not too many cores this means that a 0.25 CE and 0.30 AB, practically will serve all, but 0.25 and 1.00 on a 8 core absolutely does not, so you're left to manipulate the CE value.... set it to 1.0.

Lots been said and inserted my observations and opines over at developers [and ignored for the most part as decisions by developers and project operators were seemingly made [lower the In Progress]. The 2 options [CE/AB] getting different descriptors that are as it was put "sufficiently vague" to apply to both old and new clients.

Basically, BOINC will load up and fetch as much as it can, but for WCG, the only way to cache up is to increase the CE value. When that CE value is near 4 days, you'll be seeing very frequent High Priority [HP] processing [I understand and discussed reasons with other formal testers], and we all know that that is not in an exactly logical order [not Earlier Deadline First!]. Want to avoid that for the most part, for instance on a 7 day off-line cycle, increase CE to 7 days, see panic evolve whilst sequential calls for work are made [and end up as in my case with 24 tasks in Waiting to Run state]. When the 7 days cache are reached [work fetch stops eventually], set CE back to for instance 1 day, then HP ending.

So what will you observe with a 0.0 CE and a 1.0 AB is that work is fetched up to 1.0 days, but at WCG with less than 16 cores, 1 but no more than 15 tasks at a time. As soon as all cores are busy again, work fetch stops. Ergo, want to ensure with even the smallest outage that no core goes idle... stick to the default CE... and given the HM of being bounced up to 24 hours to reconnect on such an outage, it's low temps on most cores for the exclusive WCG crunchers.

In summary, the design is to promote low cache, almost penalize high caching, the penalty to the micro manager being the, to many, unnerving HP processing. Forget about Set and Forget, lest you're able to just let BOINC do what it is supposed to do best... without the constant looking over shoulder.

Request to knreed et all: Please consider increasing "per-call" max to e.g. 3x number of active cores in a device i.e. an octo being able to fetch 24 and not 15, possibly scaled to the duration of the science... set a higher multiplier in one call for HCC than for FAAH. Trust me... after a month of testing on an 8 core, mobile device, the current experience is less than friendly, once again requiring an advanced degree of BOINC manipulation skills. Is that what we want from the casual volunteer? Imagine the confusion when that dreaded "There's a new version" appears in the message log... could be sooner than is liked.

--//--

edit: Sorry if I repeated what you said, DA, but there will be a learning curve to take for many on this, and some tweaking once WCG gets to release BOINC server version 700 software in production.