Rattledagger
Elite Member
Some BOINC-news, 1st for the beta-test:
In windows, BOINC v2.15 is released, seti is up to 2.13.
Main boincfixes is the "Progress" staying at 100% then done & fixes "To Completion".
I've not seen any seti-2.13-wu yet, so don't know here...
Recent average credit:
"I fixed (I hope) the formula used to calculate recent average credit,
and zeroed it out everywhere (user, team, host);
it will converge to an appropriate value in the next day or two.
If you're interested, the new code follows.
-- David" (snip)
Credit:
"We're rethinking the credit system.
The BOINC philosophy is to grant credit based on how much
work actually got done (e.g. FLOPs performed by the application),
rather than how much work could have been done.
So what matters is how the application was compiled,
not how the benchmark functions (in the core client or DLL) were compiled.
What we'll probably do is leave it up to the application to allocate credit;
it can do this either by running its own benchmarks
and multiplying these by CPU time (similar to current,
except that benchmarks are in the application),
or by using estimates of how many FLOPs are used in each of its phases.
The latter would be relevant if some of the computation
is done outside the CPU chip (e.g. by a graphics coprocessor).
Also, at some point we'll fold disk storage and network
communication into the credit formula.
-- David"
Revise credit system:
"1) improve system for allocating computational credit
a) app is passed CPU benchmarks in init file
b) app passes back current credit in shmem
it can do this either by
- multiplying CPU by core client's benchmark
- running its own benchmarks, and multiplying by CPU
- using internal estimates of FLOPs
(e.g. if using graphics chip)
2) add disk storage, network transfer to credit"
(Taskbase #1039, client - priority medium, David Anderson)
Climaprediction.net report:
"During the past three days we've been meeting with Carl Christensen,
the software architect of Climateprediction.net,
to figure out how we have to extend BOINC to handle their project.
There are many challenges; their "work units" take a month or two
to finish, generate GB-size output files (only some of which is uploaded;
the rest remains resident on the client), and are sufficiently
nondeterministic that our redundant computing/validation scheme
won't work for them.
We came up with fairly simple solutions to all the challenges,
and Carl is enthusiastic about eventually switching Climateprediction
over to BOINC.
We'll try not to let this delay the launch of SETI@home/BOINC.
But it's very exciting that we're moving towards our long-term goal -
sharing the "global computer" optimally among diverse science projects.
-- David"
"I'm having a nice time here in San Fran/Berkeley, the SETI guys are great and I'm looking forward to moving CPDN to BOINC and working with everyone. I think it will be a good "merger" and about the only main thing we need is to put trickles in BOINC and maybe move the code around so that the upload server is assigned at run-completion time (right now they assign the upload server URL when a run is allocated). I have some great pics I hope to post soon of the SETI/BOINC gang and the usual vacation pictures of the Golden Gate Bridge etc 😉 " (Carl Christensen)
A little bit more from CPDN & pictures of most of the Seti-gang, follow the link...
Lastly, a Merry Christmas & a Happy New Year. 😎 :beer: :wine:
In windows, BOINC v2.15 is released, seti is up to 2.13.
Main boincfixes is the "Progress" staying at 100% then done & fixes "To Completion".
I've not seen any seti-2.13-wu yet, so don't know here...
Recent average credit:
"I fixed (I hope) the formula used to calculate recent average credit,
and zeroed it out everywhere (user, team, host);
it will converge to an appropriate value in the next day or two.
If you're interested, the new code follows.
-- David" (snip)
Credit:
"We're rethinking the credit system.
The BOINC philosophy is to grant credit based on how much
work actually got done (e.g. FLOPs performed by the application),
rather than how much work could have been done.
So what matters is how the application was compiled,
not how the benchmark functions (in the core client or DLL) were compiled.
What we'll probably do is leave it up to the application to allocate credit;
it can do this either by running its own benchmarks
and multiplying these by CPU time (similar to current,
except that benchmarks are in the application),
or by using estimates of how many FLOPs are used in each of its phases.
The latter would be relevant if some of the computation
is done outside the CPU chip (e.g. by a graphics coprocessor).
Also, at some point we'll fold disk storage and network
communication into the credit formula.
-- David"
Revise credit system:
"1) improve system for allocating computational credit
a) app is passed CPU benchmarks in init file
b) app passes back current credit in shmem
it can do this either by
- multiplying CPU by core client's benchmark
- running its own benchmarks, and multiplying by CPU
- using internal estimates of FLOPs
(e.g. if using graphics chip)
2) add disk storage, network transfer to credit"
(Taskbase #1039, client - priority medium, David Anderson)
Climaprediction.net report:
"During the past three days we've been meeting with Carl Christensen,
the software architect of Climateprediction.net,
to figure out how we have to extend BOINC to handle their project.
There are many challenges; their "work units" take a month or two
to finish, generate GB-size output files (only some of which is uploaded;
the rest remains resident on the client), and are sufficiently
nondeterministic that our redundant computing/validation scheme
won't work for them.
We came up with fairly simple solutions to all the challenges,
and Carl is enthusiastic about eventually switching Climateprediction
over to BOINC.
We'll try not to let this delay the launch of SETI@home/BOINC.
But it's very exciting that we're moving towards our long-term goal -
sharing the "global computer" optimally among diverse science projects.
-- David"
"I'm having a nice time here in San Fran/Berkeley, the SETI guys are great and I'm looking forward to moving CPDN to BOINC and working with everyone. I think it will be a good "merger" and about the only main thing we need is to put trickles in BOINC and maybe move the code around so that the upload server is assigned at run-completion time (right now they assign the upload server URL when a run is allocated). I have some great pics I hope to post soon of the SETI/BOINC gang and the usual vacation pictures of the Golden Gate Bridge etc 😉 " (Carl Christensen)
A little bit more from CPDN & pictures of most of the Seti-gang, follow the link...
Lastly, a Merry Christmas & a Happy New Year. 😎 :beer: :wine:
