Long time BOINC users will be familiar with the two settings for cache size - Connect every x.xx days (CE) and the additional days setting (AD). Previously, the recommendation has always been to keep CE close to reality, ie very low or zero if you are always connected, and to just use AD to control the cache size.

With the new BOINCs, things have changed drastically. The two values should be more or less regarded as low and high water marks. So if you have CE close to zero, your cache will almost completely drain before refilling and if you have AD quite large, you will get a lot of work in one big hit when you refill.

Since this is a test project and apps may change suddenly without prior notice and work could be canceled without prior notice, it could be quite a waste if one of these events occurred just after you had done a big top-up. So, common sense says that (if at all possible) you should keep the two values close to each other and at reasonably low values. If you wanted to keep a 1 day cache (and you used to have 0.0 + 1.0 for the two settings) , you should use something like 0.9 and 1.0 so that the cache would be kept topped up all the time without draining all the way to zero.

Here is an example of a work cache with rather undesirable settings. A big top up has recently occurred and if you go back though the history, you will see these occurring at intervals of about 6 days. In between, the cache drains to zero. This appears to be something like a 0+6 cache which I think is undesirable for this test project.

---

Cheers,
Gary.

David (Anderson) will look at GPU work fetch problems in the BOINC 7 clients, since these seem to ignore a lot of factors. I managed to get 35 OpenCL tasks in on a CE of 0.1 + AD of 1. Been running those for 3 days already. ;-)

---

Jord.

BOINC FAQ Service

They say most of your brain shuts down in cryo-sleep. All but the primitive side, the animal side. No wonder I'm still awake.

I've done a bit more experimenting with work cache under 7.0.X. For CPU tasks anyway, the CE value seems to be the low water mark and BOINC will request more CPU work when the work on hand drops below CE.

CA itself is not the high water mark as I previously suggested. The high water mark (as it was previously) is CE + CA. So for a CPU work cache to be kept at around or just above 1 day at all times, you need CE=1.0 and CA=0.1 say. This seems to prevent a big influx of tasks when a CPU work request is made. So far, those settings seem to work OK with the OpenCL app too.

---

Cheers,
Gary.