Formula Boinc Sprints 2018

[Ken g6]

Well, that's it. We wound up 2nd in league 1, and 4th overall (with SETI.USA third!) The validator issues at the end really made it a crapshoot.

 

[StefanR5R]

The last 24 hours.

Yesterday I suspected SETI.USA to do just the bare minimum to overtake us, but they did put in some real work after all. While doing so, they apparently validated some of Rechenkraft.net's work, such that RKN just so made it past Russia in the end.

 

[TennesseeTony]

Looks like our good buddies at [H] have SUSA being a bit more cautious and covering their own backsides more than boosting XtremeSystems. [H] narrowly missed GOLD in league two.

 

[ao_ika_red]

It's a bit disappointing that in last 12 hours my machines are unable to talk to Rake's server but I still appreciate Rake's admins for their effort to keep the race alive with their limited resources.

 

[Orange Kid]

RakeSearch final
Well, we ended the siege of rakesearch in 2nd for our league, 4th overall. The XS'ers and [H]ard ones in first and second in their league and overall. Big congrats to them.
Almost a month before the next sprint here, but never fear there are other races to run, Primegrid and Seti WOW.

1 SETI.USA ..............................25 ...…….10,966,818
2 TeAm AnandTech ............…..18 ………..8,612,054
3 Czech National Team ............15 ...……...2,287,970
4 Rechenkraft.net ......…......…..12 …...……1,979,794
5 Russia Team .....................…..10 ...……..1,976,435
6 Boinc@Taiwan .....................….8 ………..1,492,016
7 Planet 3DNow! ............…...…….6 ......…..1,466,672
8 Dutch Power Cows ...............….4 …….....1,025,405
9 Overclock.net...…………………. 2 ...…...….836,952
10 BOINC@AUSTRALIA .........….1 ...…...…..607,565

...a better tally

xii5ku 3,281,727
emoga 1,625,907
crashtech 1,181,569
biodoc 641,795
Howdy2u2 402,523
10esseeTony 366,201
Orange Kid 323,547
Ken_g6 248,616
Mark 146,275
Rudy Toody 122,932
GLeeM 117,636
Fardringle 85,347
Kiska 67,345
MarcelliusIkaP 37,813
zzuupp 27,534
bradtech519 13,165

...and the F1 results
July 29: Hungarian Grand Prix
1. Lewis Hamilton, Mercedes
2. Sebastian Vettel, Ferrari
3. Kimi Raikkonen, Ferrari

 

[StefanR5R]

Cross-league results, top 30:

Same, zoomed into < 2.5 M points:

league 1: continuous lines
league 2: dotted lines
league 3: dashed lines

 

[StefanR5R]

RakeSearch final

Or, not so final.
There was another stats update at the FB site:

1 SETI.USA ................................ 25 ........... 10,966,818 --> 11,238,399
2 TeAm AnandTech ................. 18 ............. 8,612,054 --> 8,885,161
3 Czech National Team ........... 15 ............. 2,287,970 --> 2,365,343

I wonder if it will go on updating.

Edit:
Checked again now; it didn't update further.
I suppose it took FB more than a few minutes to fetch stats from the project server. At 22:02:50 UTC, our score at FB was 8.6 M, at 23:02:50 UTC it was 8.8+ M. I don't know when precisely it went from the former to the latter.

 

[Orange Kid]

Or, not so final.
There was another stats update at the FB site:

1 SETI.USA ................................ 25 ........... 10,966,818 --> 11,238,399
2 TeAm AnandTech ................. 18 ............. 8,612,054 --> 8,885,161
3 Czech National Team ........... 15 ............. 2,287,970 --> 2,365,343

I wonder if it will go on updating.

Hmmmm, that's interesting

 

[Howdy]

Cannot wait until the next project FB can kill!

 

[StefanR5R]

Cannot wait until the next project FB can kill!

...or if not kill outright, zombify?!

Besides 1056 tasks still pending validation, I am left with 13 tasks listed at the web site as "in progress", although I have none left locally. They are all associated with the same host (which ran only a single client instance during the entire sprint), and are shown as sent over a time frame of 6 seconds from the first to the last of these zombie tasks.

I had the very same problem already at the Cosmology@Home sprint, both at the beginning of the sprint after I emptied buffers from Bunker Thursday, and at the end of the sprint.

Either a scheduler transaction when requesting tasks went wrong (failure to the client, but seen as success by the server), or a scheduler transaction when reporting completed tasks (successful report from the client's POV, but server-side failure).

 

[Kiska]

Either a scheduler transaction when requesting tasks went wrong (failure to the client, but seen as success by the server), or a scheduler transaction when reporting completed tasks (successful report from the client's POV, but server-side failure).

That can happen frequently if the server is overloaded. The database has assigned your host the task, but the web server has reached its php timeout and sends scheduler_reply that has been dynamically generated whether or not it is complete, so those tasks may not have made it to the reply but got assigned anyway

 

[StefanR5R]

We received our canoe badge:

http://rake.boincfast.ru/rakesearch/team_display.php?teamid=1758
http://rake.boincfast.ru/rakesearch/forum_thread.php?id=101#504

 

[ao_ika_red]

Besides 1056 tasks still pending validation, I am left with 13 tasks listed at the web site as "in progress", although I have none left locally.

Thanks to your TeAm's RakeSearch link, now I can access my own profile and I can confirm that I also has this 21 zombie tasks.

 

[[H]Coleslaw]

Looks like our good buddies at [H] have SUSA being a bit more cautious and covering their own backsides more than boosting XtremeSystems. [H] narrowly missed GOLD in league two.

That didn't stop SUSA from showing up with a dump or two for them. I think Bryan did around a 2 million point bunker drop for them.

 

[StefanR5R]

RakeSearch appear to have hired SETI.Germany's Jeeper.
They published an extensive Race report and a badge award ceremony as a boinc notice and on their front page.

 

[ao_ika_red]

RakeSearch appear to have hired SETI.Germany's Jeeper.
They published an extensive Race report and a badge award ceremony as a boinc notice and on their front page.

That's very Jeeper indeed.

 

[Rudy Toody]

RakeSearch appear to have hired SETI.Germany's Jeeper.
They published an extensive Race report and a badge award ceremony as a boinc notice and on their front page.

That is a hoot!

 

[zzuupp]

That is a hoot!

Sorry, I couldn't resist.

 

[biodoc]

The sprint project is yoyo@home.

 

[zzuupp]

I'm feeling too lazy to do research. So, I'll just ask, "Which is the best subproject for this?"

I've got my defaults of a few I've been working on, but they aren't necessarily 'sprint worthy'.

 

[ao_ika_red]

I'm feeling too lazy to do research. So, I'll just ask, "Which is the best subproject for this?"

I've got my defaults of a few I've been working on, but they aren't necessarily 'sprint worthy'.

Cruncher OGR, ECM, and PerfectCuboid for now. But I'll change it later to PerfectCuboid only.

 

[StefanR5R]

I'm feeling too lazy to do research. So, I'll just ask, "Which is the best subproject for this?"

For the subprojects which were live in spring 2017, check out the discussion on the 1st 2017 sprint:
https://forums.anandtech.com/threads/formulaboinc.2500571/
(especially page 2)

Also, it was sprint project of Pentathlon 2017:
https://forums.anandtech.com/threads/8th-annual-boinc-pentathlon.2503356/page-14#post-38886853

 

[StefanR5R]

Looks like yet another project with bogus crediting.
https://www.rechenkraft.net/forum/viewtopic.php?f=56&t=11559&start=192

 

[ao_ika_red]

Looks like yet another project with bogus crediting.
https://www.rechenkraft.net/forum/viewtopic.php?f=56&t=11559&start=192

Atom beat Haswell, yeah right.

 

[StefanR5R]

The RAM requirement of ECM tasks can be capped by a command line switch, to be specified in app_config.xml. Example syntax:

-maxmem 2000​

for 2 GB. (Needs a proper app_version block and cmdline tags around it.) If the particular task needed more than is allowed by the command line switch, it will still run, but its run time will increase.
(source in German, see 3rd post. Alas the syntax gets lost in machine translation. The way how boinc checks for memory consumption is also discussed in this thread.)

Edit 2,
I only remember tasks with up to ~2 GB RAM usage, from the few and brief occasions that I ran yoyo@home. But according to the link above, there had been batches of WUs with up to 10 GB RAM requirement.

Furthermore, I understand from this thread: The server may send you tasks which individually may need as much RAM as is configured in BOINC in the "Disk and memory" options. While the task runs and is allocating memory, the client should notice and either suspend other tasks, or this task, into "waiting for memory" state. But if the client is not quick enough to check, memory allocations will fail and tasks will crash.

--------
Edit 1,
concerned parties have opened two threads at the yoyo numbercrunching forum, in order to alert the admin about the impending server load:
https://www.rechenkraft.net/forum/viewtopic.php?f=57&t=16976
https://www.rechenkraft.net/forum/viewtopic.php?f=57&t=16977

However, since RKN, including yoyo, are actively participating in FB sprints, he noticed anyway in the morning.
https://www.rechenkraft.net/forum/viewtopic.php?f=92&t=16857&start=408#p175353