Making connect-to-network setting more user-friendly
4 Aug 2005 10:20:57 UTC
Topic 189660
(moderation:
As far as I can tell, the number of work units queued is determined by the processing power of the host and the connect-to-network time. Given a connection interval of 1 day, work units worth 1 day of CPU time for that host are downloaded.
This works perfectly well for hosts that run 24 hours a day. Unfortunately, it happens that many hosts crunch on multiple projects, or don't crunch 100% of the time. What happens then is that the connect-to-network interval needs to be set proportionally less.
First of all, this is not intuitive. There seem to be many cases of this setting being set to high, as evidenced by threads in the forums to this effect.
Secondly, it is potentially inconvenient for dial-up users to use a smaller connection interval than is desired.
It seems it would be beneficial to make the connection interval somewhat independant of the queue length. Perhaps the addition of a 'duty cycle' field would suffice, where the user estimates the percentage of time that host will be crunching 'Einstein' work units.
Anyhow, I just mean to make this anomaly known. Ideas for a solution or the merits of finding a solution should be discussed.
Making connect-to-network setting more user-friendly
)
Doesn't the resource share already indicate how much CPU the user wants Einstein to have or intends to give it?
If the download was proportioned to the resource shares, it shouldn't be as big a problem.
RE: Doesn't the resource
)
-Edit: I should read the post before replying (multiple times if necessary)
Making it proportional to the resource share would surely solve a predominant number of cases. I don't know that there are that many people that choose only to crunch for a few hours per day, so that would be a good remedy.
I mean to have all downloads
)
I mean to have all downloads (however many hours, days) to be proportional to the resource share.
The client calculates a few
)
The client calculates a few variables that track how the client is using its time. On time - the fraction of time that the machine spends on. Up Time - the fraction of on time that BOINC spends running. CPU effeciency (new for 4.70) the fraction of time that BOINC apps get the CPU of the time that BOINC spends running. The Resource shares for all projects. In 4.2x the calculation was moved from the client to the server. I would like to move it back. Some servers are still using code old enough that it doesn't do the calculation on the server, but the server is counting on the client doing the calculation. This is what leads some projects to hand out way too much work when asked. All of these values take some time to stabilize at a new value, and this can lead new computers to get too much work the first time.
BOINC WIKI