Bok
09-09-2005, 09:57 AM
From their announcements..
Production continues of a second batch of 250,000 million-turn jobs. The first such batch was successfully completed on August 26th after running for about 2 months. Each batch is actually being submitted in 8-10 sub-batches of around 25,000 jobs and each job creates approximately 5 Results due to the BOINC redundancy factors chosen. Each million-turn WU takes about 5 hours on a typical modern CPU, but we use more generous time estimates to accommodate the inaccurate benchmark in the boinc client. A small percentage of WU's terminate very rapidly in the case where the particles are lost due to beam instability - which is very interesting for us if not for the donor's credit - sorry!
An analysis is performed after each sub-batch. It is important to the project that all jobs in a sub-batch are completed before each analysis can proceed - for this reason we use fairly aggressive deadline settings and adjust them depending on observed turnaround. Deadlines are currently around 5 days. As experienced crunchers have noted, this should discourage "hoarding" of WU's in large caches, a practice which adversely affects our completion times.
The actual process of job submission is driven by a very small number of machine physicists, and irregularities are to be expected. Yes, sometimes we run out of work and sometimes we have too much. Please be tolerant, and please try and support other BOINC projects when we are out of work. We may be analysing (or we may be tired or on holiday).
The work being done is a serious contribution to the commissioning of the LHC accelerator, and the amount of work in prospect will increase as commissioning proceeds. This is because the beam simulations initially use "nominal" machine parameters, but as real machine components are installed the true measured parameters are introduced and the results become more critical. Currently we are using "Version 6.4" of the LHC machine configuration: the updates to each later machine version will require a new series of runs. Version 6.5 is expected shortly.
We removed the freeze on new donors about 2 months ago and are receiving around 3000 new active subscribers per month. We are pretty happy with this situation and have no server or bandwidth problems for now.
Thanks once again to all our supporting crunchers and to all the experienced people out there who answer many (if not all) of the questions posed by newcomers.
LHC@Home Team
Production continues of a second batch of 250,000 million-turn jobs. The first such batch was successfully completed on August 26th after running for about 2 months. Each batch is actually being submitted in 8-10 sub-batches of around 25,000 jobs and each job creates approximately 5 Results due to the BOINC redundancy factors chosen. Each million-turn WU takes about 5 hours on a typical modern CPU, but we use more generous time estimates to accommodate the inaccurate benchmark in the boinc client. A small percentage of WU's terminate very rapidly in the case where the particles are lost due to beam instability - which is very interesting for us if not for the donor's credit - sorry!
An analysis is performed after each sub-batch. It is important to the project that all jobs in a sub-batch are completed before each analysis can proceed - for this reason we use fairly aggressive deadline settings and adjust them depending on observed turnaround. Deadlines are currently around 5 days. As experienced crunchers have noted, this should discourage "hoarding" of WU's in large caches, a practice which adversely affects our completion times.
The actual process of job submission is driven by a very small number of machine physicists, and irregularities are to be expected. Yes, sometimes we run out of work and sometimes we have too much. Please be tolerant, and please try and support other BOINC projects when we are out of work. We may be analysing (or we may be tired or on holiday).
The work being done is a serious contribution to the commissioning of the LHC accelerator, and the amount of work in prospect will increase as commissioning proceeds. This is because the beam simulations initially use "nominal" machine parameters, but as real machine components are installed the true measured parameters are introduced and the results become more critical. Currently we are using "Version 6.4" of the LHC machine configuration: the updates to each later machine version will require a new series of runs. Version 6.5 is expected shortly.
We removed the freeze on new donors about 2 months ago and are receiving around 3000 new active subscribers per month. We are pretty happy with this situation and have no server or bandwidth problems for now.
Thanks once again to all our supporting crunchers and to all the experienced people out there who answer many (if not all) of the questions posed by newcomers.
LHC@Home Team