Few days ago I started an offline discussion and got this reply from the WCG chief architect, in snip:

A. Reliability is no longer based on a device basis, but on a device/science app combo.

B. The device must have met 3 criteria for a specific science.

1. Done 5 Valid results consecutively (not returned, but validated).
2. The maximum quota must be applying (120 per core per day). Errors reduce that quota immediately to a lower number.
3. Your results must get reported (the ready to report bit), on average within 48 hours.

--//--

P.S. Being reliable does not mean you get them. Depends on how many reliable devices hav been rated for each of the individual sciences, and how many errors need to be fixed... if few, you wont see them either or rarely. ATM I'm getting them frequently on octo and quad for CFSW, but at 60-70 per day without error, that means loads of validation and re-confirmation the hosts are fit to be in the ''second opinion'' league. Whilst, long moons ago knreed was suggesting a congratulations message to reliable hosts (we did not discuss a retraction message ;P)... who knows one day, one spare hour and we'll get it... it's one of those ''nice to have, but not essential''.

I have long since met those criteria, except that like Bugg, I am not sure what "120 per core per day" means. As I pointed out above, the only errors I get are on HPF2, and then only maybe maybe one per 20 work units or more. I think their servers are not up to speed in some sense. Maybe they have multiple servers each with their own set of reliability data?

A computer below the max daily quota has never been "reliable", but since this quota is doubled for each "success"-report having a few errors doesn't really matter, since often a single report is enough to meet this requirement again.

The turnaround-time is also easy to meet again, since it's a weighted average with 30% weight for new tasks, but it's slowed-down somewhat due to this being validated results.

It looks like the old-style global quota may still be in place, even the info about this quota has been removed from the computer-info-page, while the per-application quota is shown on a separate page together with some other info.


Especially since the per-application quota starts as low as 2 per core, and this only increases by 1 for each Validated result, keeping track of the per-application quota is more important than the "global" quota that normally will be at 120.

"used 20 of daily quota"... not sure on the interpretation here, since is it "downloaded 80" (if quad-core), or "downloaded 20"? If the latter, it means computer can still grab 460 more wu's this "day".

BTW, neither of the quotas has any history-function, so there's no way to know if "last 5 valid" came 1 month ago, or happened due to 5 validated results passing-through the validator 5 minutes after one invalid.

Got a little surprise here then looked-it up, while quota before was reset at midnight, this gave an extra spike of work-request & downloads following midnight and the quota-reset was therefore changed. Each host-id now has a random-generated "reset-time", and this is uniformly distributed over the full 24 hours.

Example, someone running 2 computers can have one computer resetting the quota at 10:45, while the other computer resets quota at 16:36.

Meaning, "midnight" doesn't matter any longer, and there's no way to know the time-of-day any individual computer will reset the quota.

FWIW, both my machines run 24/7 and always complete the work units in the required time, for at least the last year.

Is "trusted", and "reliable" as used in the context of WCG's server's assessment of client machines connecting to it -- mean the same thing? I recall there is a difference.
;