2013-02-03

Muon1, a 2012 Retrospective

A little late, but I thought I’d bring you the breakdown of results for Muon1 through 2012. Better late than never though, so here are the highs and lows of 2012 for the Muon1 Distributed Particle Accelerator Project.

To start with, let’s start with the stats. The weekly stats continue to be published every Tuesday, and the start and end points this year for this piece are 3 January 2012 to 1 January 2013.

They were
January 3 2012
30,162 Trillion Particle timesteps overall (+193T over the past week)
Linac900Ext10tc2 is at 3.772285%, found by a BOINC Wrapper (yoyo@home) user, an increase of 0.000947% from last week.
Linac900Ext9X is at 1.734988%, found by [SG]PIT-V, an increase of 0.080691% over the last week.

January 1 2013
38,934 Trillion Particle timesteps overall (+137T over the past week)
Linac900Ext9Xc2 is at 4.022552% found by [ARS]whizbang an increase of 0.000134% over the past week
Linac900Ext8Xc2 is at 2.041505% found by [ARS]whizbang an increase of 0.024396% over the last week
Linac900Ext6Xc2_nosample is at 3.891581% found by [ARS]Dave Peachey an increase of 0.051066% over the last week

MPTS

MPTS-wise, we’ve gone from 30,162 Trillion, to 38,934 Trillion – an increase of 8,772 Trillion. That works out to an average of 24.1Trillion a day. To put it in perspective, that’s simulating a particle at lightspeed for 72,495,867 km (45,046,843 miles) – half the distance from the Earth to the Sun, 4 light minutes – every single day.
MPTS over time (click to enlarge)

It also shows the incredible increases in both computing power, and the project’s userbase, as almost one quarter of the work done in the last 12 years was done in 2012.

Lattices

No lattices that were running in at the start of 2012, were still active at the end of the year. While they all still accept data (as indeed do all past lattices) they’re no longer a primary focus.

February 10 saw the main change, as both Linax900Ext10c2 and Linac900Ext9X were closed.
Replacing them were Linac900Ext9Xc2 and Linac 900Ext7Xc2.

A month later (March 16), a new experiment, in the form of Linac900Ext6Xc2_nosample was started. Then in October (October 22) Linac900Ext7Xc2 was closed, and replaced with Linac900Ext8Xc2.
Lattice Yields over time (click to enlarge)
lattice yields and MPTS over 2012 (click to enlarge)

Major Events

In March, to coincide with the launch of the first _nosample lattice, the samplefiles were changed from being updated every 3 hours, to once a week. It was part of an experiment that has been ongoing all year on how to improve the optimizer. This ended in December, as the slower updates were found to not help the clustering approach much.

In May, the stats server went down, and was unavailable until early June. With most distributed projects, the loss of one of the central servers would cause significant problems, but because of the decentralized nature of the genetic algorithm, performance was unaffected.

August 31st, we hit 75 million simulations returned. It’s quite a milestone, and one we should have celebrated more, but it was conference time, and we just couldn’t.

Other Events

The ‘best design’ series of videos was brought up to date in May, while 7Xc2, which finished after that point, had its video released on the 12 December (after a 6 week wait for last minute results)

Meanwhile, the BOINC users are anonymous no more! From mid-March, all Yoyo@home wrapper users had their usernames parsed once more, meaning they were no-longer identified as ‘Yoyo@home Boinc wraper’. They can still be identified via the (BOINC) after their name, but they have a name again now. Once again, thanks to Yoyo for another year of supporting Muon1.

2013

2013 is already underway, and some of the upcoming things are more milestones, and more new, exciting lattices. There may even be some new client versions, and who new design paths – but who can tell!

In the meantime, keep crunching everyone, and don't forget to follow us on Facebook, Twitter or Google+

K`Tetch

No comments:

Post a Comment

Post a Comment