There's a problem with the way the Scheduler handles the reply to the host when a request receives 'DB-less Buzz Off' from the scheduler.

In a stock BOINC scheduler, all scheduler requests generate a host record lookup in order to update any global preference changes (regardless of where they originated from) as well as send updated credit, venue, and other info back (project specific) to the host.

As the reader may recall, MW has modified the scheduler so when the project is under high load or other atypical situations, a scheduler request from a host can be responded to and completed without having to make the host record lookup if certain conditions apply.

The problem is that the reply is of a 'generic' format, and may not necessarily represent what the actual combined preferences settings are for the host making the request.

Specifically, I have observed that if you have set your host to use a venue other than the default one and you get a 'DB-less' reply from the scheduler, it will switch the venue of the host from the one which was set in the host preferences to whatever you have set for the default venue. This condition will persist until such time that the host gets a 'standard' reply back from the scheduler.

It would seem the most straight forward solution to this would be for the scheduler to suppress sending venue preference data back to the host in cases where the host record is not actually looked up. The reason is an arbitrary switch to a venue the user didn't select can result in significant changes in the way the CC runs the work stream. In the best case, this results in unnecessary Resource Share breaks, and in the worst case could even result in blown deadlines for some tasks given the right set of preference parameters (since MW is such a tight deadline project).

Alinator