Hello all.
Just a brief report to tell you how are things going on.
The fix in the credit granting seems to work, now RAC and total credit values in statistics are little by little adapting and showing the real performance of each host.
During the weekend I raised the WUs per host limit to 2. That allowed us to reach the number of 4500 results per hour.
So far, he have computed half of the WU and there is no promising solution found yet.
On late Sunday I set again the limit to 1 because the last WUs are requiring slightly more than 2 GB each one and that kills computers with only 4 GB.
Also, I raised the deadline time from 24 to 48 hours.
What I have found is that our rate of results/hour has dropped to some 1600, which is less than a half. I wonder if the deadline increase has something to do with this.
I have done some SQL queries to the database, and found the following results
This is more or less in line with the poll done some days ago: 58% of the hosts available has less than 6GB, and contributes with 41% of the credits. On the other side, 42% of the hosts has 6GB or more and contributes with 59% of the credit. It is not shown in the table, but there are 3 hosts with more than 24GB and 2 hosts with more than 32GB (!)
Anyway, credit values are still not completely representative, because of the issue with Linux and credits claimed we had last week that makes Linux high-performance machines have more credits than deserved.
Hope you find those numbers interesting.
/ 
I have one host with 4GB and three with <4GB if that helps the estimate any.
/ 
Since you change the deadline, the WU wait and wait and wait before starting. In my opinion, a deadline of 24 hours in good for a 4 minutes WU (even 12 hours is good).
/ Since you change the deadline, the WU wait and wait and wait before starting. In my opinion, a deadline of 24 hours in good for a 4 minutes WU (even 12 hours is good).
I'm lowering now the deadline to 12 hours for a while, please check if there is any improvement. For the record, we processed 48760 results in the last 24 hours and 2436 results in the last hour before the change.
/ Yes, it works perfectly. As soon as a WU is finished, I get a new one which starts immediately :-)
/ I'm one of the 24 hosts with 8GB to 12GB, though I've been contemplating installing additional RAM for my BOINC machine if it becomes necessary.
I can't speak for anyone else, but a 12-hour deadline is fine for me. I just raised Neurona's resource share to 125 from the default 100, to ensure my machine is 'always' doing these WUs in time.
/ and 2 hosts with more than 32GB (!)
Hmmm. Sorry. If I've messed your numbers up, I can move those two hosts elsewhere. :-)
I'm surprised after your call for larger RAM footprint machines, that's all that made it here. We'll have to see if we can't do better for you than that as your WU sizes continue to rise.
/ and 2 hosts with more than 32GB (!)
Hmmm. Sorry. If I've messed your numbers up, I can move those two hosts elsewhere. :-)
I'm surprised after your call for larger RAM footprint machines, that's all that made it here. We'll have to see if we can't do better for you than that as your WU sizes continue to rise.
Move them? No, it is not necessary, I think I can live with it ;-)
No, seriously, you have in these two hosts more RAM and CPU power than I have in the 40 computers my degree students use for its first steps in programming, so it is quite impressive.
I hope we will be able to attract more people with more horsepower as the project develops. I'm pretty happy with what we have achieved so far.
/ Since you change the deadline, the WU wait and wait and wait before starting. In my opinion, a deadline of 24 hours in good for a 4 minutes WU (even 12 hours is good).
I'm lowering now the deadline to 12 hours for a while, please check if there is any improvement. For the record, we processed 48760 results in the last 24 hours and 2436 results in the last hour before the change.
Volunteers can force completed tasks to report immediately by using the report_results_immediately option in a cc_config.xml file in the BOINC data directory.
<options>
<report_results_immediately>1</report_results_immediately>
</options>
</cc_config>
This will also cause hosts to get another task immediately. This should increase Neurona's work rate if hosts are waiting many minutes to download another task.
/ Since you change the deadline, the WU wait and wait and wait before starting. In my opinion, a deadline of 24 hours in good for a 4 minutes WU (even 12 hours is good).
I'm lowering now the deadline to 12 hours for a while, please check if there is any improvement. For the record, we processed 48760 results in the last 24 hours and 2436 results in the last hour before the change.
Volunteers can force completed tasks to report immediately by using the report_results_immediately option in a cc_config.xml file in the BOINC data directory.
<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
</options>
</cc_config>
This will also cause hosts to get another task immediately. This should increase Neurona's work rate if hosts are waiting many minutes to download another task.
Very interesting, I'm putting this in a dedicated thread.
/