It doesn't matter. As I said in my previous post, you should set the connect-to-network time according to the queue length you want. With your hosts, if you were running them 24 hours per day, 7 days a week a queue length of 8 would be okay, because you would manage to process 8 in a week. That would dictate a setting of 2.5 days.
But since you are sharing, I don't think it should be above 3 (given a 30% share). That means the setting should be no more than 1 day. Any more than that and you cause them to time out, potentially wasting resources.
It doesn't matter. As I said in my previous post, you should set the connect-to-network time according to the queue length you want. With your hosts, if you were running them 24 hours per day, 7 days a week a queue length of 8 would be okay, because you would manage to process 8 in a week. That would dictate a setting of 2.5 days.
But since you are sharing, I don't think it should be above 3 (given a 30% share). That means the setting should be no more than 1 day. Any more than that and you cause them to time out, potentially wasting resources.
You Know In a perfect world all the Boinc Project servers wouldn't crash all the time and people who just want to help don't have to be a rocket scientist.
I set my project to suit my needs, and it ain't 1 day~!
You Know In a perfect world all the Boinc Project servers wouldn't crash all the time and people who just want to help don't have to be a rocket scientist.
Who ever told you the world was perfect?
I am just relaying how it seems to work. The connect-to-network time is used to determine how long the queue is, but it assumes that the host will be running all the time. If you have a lower duty cycle or crunch on multiple projects, it downloads too much.
I agree that you should be able to set the connection interval to whatever you like, but 'should' and 'is' are very different things.
Here is a work unit which my host completed. There have been two timeouts. Notice that this unit has been delayed, it was sent on 25 July but it still isn't complete. If host 17168 doesn't submit a valid result in time, the two completed results will be wasted.
I just meant that if you make a compromise the system can work more smoothly. It's not perfect, but nobody said it was. With a shorter connection time it will actually download less, which should be something good, considering that you are on dial-up.
I am going to post an item in the Wish List forum about adding a 'duty cycle' option to BOINC, so that the connection interval is more user-friendly.
Here is a work unit which my host completed. There have been two timeouts. Notice that this unit has been delayed, it was sent on 25 July but it still isn't complete. If host 17168 doesn't submit a valid result in time, the two completed results will be wasted.
Just a small possible clarification?
My understanding is that there is no deadline for a quorum to be reached. If 17168 doesn't return by its deadline, another box will be sent the WU. This is hinted at in the last line of the detail for the WU. And there is no time limit for how many more times this can occur. Whenever a 3rd result is completed, and if it agrees with the first two, all three machines will get credit.
IOW, no completed WU is ever wasted, unless the verification process finds it is in fact invalid. If it turns out that the WU is "outvoted" 2 to 1 say.
My understanding is that there is no deadline for a quorum to be reached. If 17168 doesn't return by its deadline, another box will be sent the WU. This is hinted at in the last line of the detail for the WU. And there is no time limit for how many more times this can occur. Whenever a 3rd result is completed, and if it agrees with the first two, all three machines will get credit.
IOW, no completed WU is ever wasted, unless the verification process finds it is in fact invalid. If it turns out that the WU is "outvoted" 2 to 1 say.
Well for whatever reason, that last result line is marked as 'unsent'. I know that if there are too many errors the unit is marked as such and aborted. I suppose the administrators feed it back into the system again.
Usually a fourth host is allocated, unless 3 valid results are in. If the fourth host fails then, no further action is taken. However, in this case no fourth host was allocated. I can only imagine that that is because the number of errors has reached the threshold. I figure that if that last pending result times out, the work unit will be marked as having too many errors.
It is true that I don't really know if those completed results are kept, or if it starts anew when it is reentered. At the very least, the work unit is delayed. I don't know how much of an issue that is for the scientists/administrators, but they do allocate the fourth host to ensure the results come in more regularly, so I assume they care that they do come in regularly. With the amount of overqueueing going on, the attempts at keeping them regular are thwarted.
Being on a dialup, I would think one would not want to download more than one has to.
I set my project to suit my needs, and it ain't 1 day~!
Then please don't complain when your client goes into EDF, and then won't download work for a week...
Well said Heffed. I agree completely.
Now can we all get back to our normal lives and let this thread die as it is obvious that the original poster isn't really interested in our advice anyway?
Now can we all get back to our normal lives and let this thread die as it is obvious that the original poster isn't really interested in our advice anyway?
If our normal lives are having 2 timeouts per work unit on average then: yeah, I suppose. It just seems not to be a good situation.
I set my project to suit my needs, and it ain't 1 day~!
Then please don't complain when your client goes into EDF, and then won't download work for a week...
Well said Heffed. I agree completely.
Now can we all get back to our normal lives and let this thread die as it is obvious that the original poster isn't really interested in our advice anyway?
Everyone's advise is important, but just because they don't do it your way, don't get upset. My complaint was I updated Boinc from 4.25 to 4.45 and there wasn't any work being downloaded for certain projects. Now that I know how & why It works, I will create a work buffer that suits """MY"""" needs. Right now I am experimenting with 2 days instead of 3 days. I appreciate everyone's help.
I set my project to suit my needs, and it ain't 1 day~!
Then please don't complain when your client goes into EDF, and then won't download work for a week...
Well said Heffed. I agree completely.
Now can we all get back to our normal lives and let this thread die as it is obvious that the original poster isn't really interested in our advice anyway?
Everyone's advise is important, but just because they don't do it your way, don't get upset. My complaint was I updated Boinc from 4.25 to 4.45 and there wasn't any work being downloaded for certain projects. Now that I know how & why It works, I will create a work buffer that suits """MY"""" needs. Right now I am experimenting with 2 days instead of 3 days. I appreciate everyone's help.
First, the amount of work downloaded is partly based on the amount of time that the computer spends on. (on Time * Up time * resource fraction * CPU efficiency). CPU efficiency was added in 4.72 (so you won't have it yet). However, it takes a little time to stabilize on these values.
Second, some of the design decisions were made with modem users in mind. The most important thing was to try to get enough work on their machines to outlast their disconnected time. The second most important thing was making reports by deadlines. Third was a mix of work, and after that were things as honoring the resource share in the short term.
RE: I use dial up! It
)
It doesn't matter. As I said in my previous post, you should set the connect-to-network time according to the queue length you want. With your hosts, if you were running them 24 hours per day, 7 days a week a queue length of 8 would be okay, because you would manage to process 8 in a week. That would dictate a setting of 2.5 days.
But since you are sharing, I don't think it should be above 3 (given a 30% share). That means the setting should be no more than 1 day. Any more than that and you cause them to time out, potentially wasting resources.
RE: RE: I use dial
)
You Know In a perfect world all the Boinc Project servers wouldn't crash all the time and people who just want to help don't have to be a rocket scientist.
I set my project to suit my needs, and it ain't 1 day~!
RE: You Know In a perfect
)
Who ever told you the world was perfect?
I am just relaying how it seems to work. The connect-to-network time is used to determine how long the queue is, but it assumes that the host will be running all the time. If you have a lower duty cycle or crunch on multiple projects, it downloads too much.
I agree that you should be able to set the connection interval to whatever you like, but 'should' and 'is' are very different things.
Here is a work unit which my host completed. There have been two timeouts. Notice that this unit has been delayed, it was sent on 25 July but it still isn't complete. If host 17168 doesn't submit a valid result in time, the two completed results will be wasted.
I just meant that if you make a compromise the system can work more smoothly. It's not perfect, but nobody said it was. With a shorter connection time it will actually download less, which should be something good, considering that you are on dial-up.
I am going to post an item in the Wish List forum about adding a 'duty cycle' option to BOINC, so that the connection interval is more user-friendly.
RE: Here is a work unit
)
Just a small possible clarification?
My understanding is that there is no deadline for a quorum to be reached. If 17168 doesn't return by its deadline, another box will be sent the WU. This is hinted at in the last line of the detail for the WU. And there is no time limit for how many more times this can occur. Whenever a 3rd result is completed, and if it agrees with the first two, all three machines will get credit.
IOW, no completed WU is ever wasted, unless the verification process finds it is in fact invalid. If it turns out that the WU is "outvoted" 2 to 1 say.
Ken
RE: My understanding is
)
Well for whatever reason, that last result line is marked as 'unsent'. I know that if there are too many errors the unit is marked as such and aborted. I suppose the administrators feed it back into the system again.
Usually a fourth host is allocated, unless 3 valid results are in. If the fourth host fails then, no further action is taken. However, in this case no fourth host was allocated. I can only imagine that that is because the number of errors has reached the threshold. I figure that if that last pending result times out, the work unit will be marked as having too many errors.
It is true that I don't really know if those completed results are kept, or if it starts anew when it is reentered. At the very least, the work unit is delayed. I don't know how much of an issue that is for the scientists/administrators, but they do allocate the fourth host to ensure the results come in more regularly, so I assume they care that they do come in regularly. With the amount of overqueueing going on, the attempts at keeping them regular are thwarted.
Being on a dialup, I would think one would not want to download more than one has to.
RE: I set my project to
)
Then please don't complain when your client goes into EDF, and then won't download work for a week...
RE: RE: I set my project
)
Well said Heffed. I agree completely.
Now can we all get back to our normal lives and let this thread die as it is obvious that the original poster isn't really interested in our advice anyway?
Cheers,
Gary.
RE: Now can we all get back
)
If our normal lives are having 2 timeouts per work unit on average then: yeah, I suppose. It just seems not to be a good situation.
RE: RE: RE: I set my
)
Everyone's advise is important, but just because they don't do it your way, don't get upset. My complaint was I updated Boinc from 4.25 to 4.45 and there wasn't any work being downloaded for certain projects. Now that I know how & why It works, I will create a work buffer that suits """MY"""" needs. Right now I am experimenting with 2 days instead of 3 days. I appreciate everyone's help.
RE: RE: RE: RE: I set
)
First, the amount of work downloaded is partly based on the amount of time that the computer spends on. (on Time * Up time * resource fraction * CPU efficiency). CPU efficiency was added in 4.72 (so you won't have it yet). However, it takes a little time to stabilize on these values.
Second, some of the design decisions were made with modem users in mind. The most important thing was to try to get enough work on their machines to outlast their disconnected time. The second most important thing was making reports by deadlines. Third was a mix of work, and after that were things as honoring the resource share in the short term.
BOINC WIKI