Formula Boinc Sprints 2018

[ao_ika_red]

Hello all.
Good to be back.
I was only able to join the sprint on Sunday but oh my, I got almost no tasks. I hope Sebastien will pick better project on future races so normal bloke like me doesn't have to do Stefan-level of tinkering just to get more tasks.

 

[StefanR5R]

Hello all.
Good to be back.

Indeed!

I was only able to join the sprint on Sunday but oh my, I got almost no tasks.

I had a look at my logs and tasks lists now.

• Yesterday until 13:36 UTC (8.5 hours before the sprint ended), I received barely enough tasks to keep my equipment occupied (which when running camb_boinc2docker is bottlenecked by upload bandwidth, not by processor capacity, as mentioned).
• My work buffers from before this point in time lasted 1.5 hours (again, determined by my upload capacity).
  
• After that time, I only received 7 more tasks at 14:23 UTC and 1 more task at 16:00 UTC.
• From then on, including 8 hours after the end of the sprint, I received no tasks anymore. (I kept requesting tasks this long after the sprint, to no avail.) Now, 11 hours after the end, I tried with one client, and still no more tasks.
• I saw the global number of tasks in progress on C@H's server status page steadily declining.

To summarize:
It looks as if camb_boinc2docker_work_generator (although is still shown as "running") finally used up its entire parameter set for new tasks several hours before the sprint concluded, and needs to be set up newly when the admin returns.

I hope Sebastien will pick better project on future races so normal bloke like me doesn't have to do Stefan-level of tinkering just to get more tasks.

Re tinkering:
Much of what I do is prompted by the above-average core count of my hosts. If I didn't do anything, they could be handicapped in sprints on quorum-2 projects, and would be pretty much useless in sprints on projects with slow work generator (such as Cosmo).

Re projects to pick:
It is clear from a few sprints last year, beginning with Asteroids@home in July 2017, and several of the sprints this year, that the concept of initiating sprints without notification of (and cooperation by) project administrators is worse than flawed.

There are 10 more sprints this year. Looking at the list of possible sprint projects, I doubt that there are 10 projects in this list that would work for the current mode of FB sprints.

• I am seeing one of which I know that it can carry a surprise sprint (PrimeGrid).
• OTOH I see two which are no longer operating (dDM, POGS), two inactive (DENIS, DrugDiscovery), one which doesn't have a Windows application (Wanless2), one which never has tasks for more than a few hours per month (XANSONS), and another one which AFAIK is often out of tasks for long periods too (BURP).
• I am unclear about the other 15 projects (Acoustics, CSG, Collatz, Einstein, GPUGrid, LHC, Milkyway, NumberFields, ODLK, RakeSearch, SRBase, TN-Grid, Universe, YAFU, Yoyo). I ran only some in the past months and noticed rather slow work generators at some of them outside any contest.

 

[biodoc]

There's a license issue for the windows app on GPUGrid so currently it's linux only.

 

[ao_ika_red]

Indeed!

You don't know how I missed this forum and especially you all DCers.

To summarize:
It looks as if camb_boinc2docker_work_generator (although is still shown as "running") finally used up its entire parameter set for new tasks several hours before the sprint concluded, and needs to be set up newly when the admin returns.

Just in time when I decided to join the sprint. I even chose all apps but still no tasks after 1500 UTC (I fell asleep after that, unfortunately). But, thankfully we already passed that project now.

It is clear from a few sprints last year, beginning with Asteroids@home in July 2017, and several of the sprints this year, that the concept of initiating sprints without notification of (and cooperation by) project administrators is worse than flawed.

This is probably the most polite way of saying, "Hey Sebastien, get your act together!"

I am unclear about the other 15 projects (Acoustics, CSG, Collatz, Einstein, GPUGrid, LHC, Milkyway, NumberFields, ODLK, RakeSearch, SRBase, TN-Grid, Universe, YAFU, Yoyo). I ran only some in the past months and noticed rather slow work generators at some of them outside any contest.

I'll try to remember some of them, feel free to correct me if I'm wrong.

• CSG: quite solid performer but has lengthy tasks completion duration
• Einstein: also solid performer as I'm a regular cruncher and almost never encounter any issue except for planned maintenance
• LHC: a bit sensitive to some ISP
• Milkyway: has task limit so it needs some workaround
• RakeSearch: bulletproof for first 2 days but slowly given up
• Universe: Linux friendly but not so friendly to heavyweight client (I'm looking at Xeon and TR users)
• YAFU: SNAFU

 

[StefanR5R]

There's a license issue for the windows app on GPUGrid so currently it's linux only.

Looks like their recurring license expiry problem.

And if problems such as Cosmo's camb_legacy validator being off remain undetected in Sébastien's workflow, this GPUGrid issue could easily slip by him too.

• CSG: quite solid performer but has lengthy tasks completion duration
• Einstein: also solid performer as I'm a regular cruncher and almost never encounter any issue except for planned maintenance
• LHC: a bit sensitive to some ISP
• Milkyway: has task limit so it needs some workaround
• RakeSearch: bulletproof for first 2 days but slowly given up
• Universe: Linux friendly but not so friendly to heavyweight client (I'm looking at Xeon and TR users)
• YAFU: SNAFU

Hmm, Einstein, LHC, and Yoyo were up in 2017's Pentathlon, and NumberFields and Universe in 2018's. Pentathlon load is a different ballgame than a Formula BOINC sprint, but (1.) FB sprinters seem to have become more active this year, combined with massively increased throughput from Gridcoin, and (2.) project admins specifically prepare well in advance for Pentathlon. See Asteroids, which almost - but not entirely - kept up with demand during Pentathlon 2018, but did not have enough work in 2017's and 2018's FB sprints.

The very long task run times of CSG, plus validation lag, make it a bad choice for a three day sprint indeed. On the other hand, the long run times mean that the work generator, database etc. have a better chance to keep up with demand during a sprint.

 

[ao_ika_red]

FB sprinters seem to have become more active this year, combined with massively increased throughput from Gridcoin

A quick glance on FB stats will tell us that healthy (but sometimes a bit salty) battles are raging all over three leagues, so yeah, surely it indicates there are more users than ever.
But I'm indeed quite surprised with GC performance as this virtual money value doesn't seem as good as in 2017 / early 2018.

 

[StefanR5R]

I'm indeed quite surprised with GC performance as this virtual money value doesn't seem as good as in 2017 / early 2018.

Gridcoin, Foldingcoin etc. may not decline as soon as some other obscure coins. They differ from several other proof-of-work based coins by being more ASIC resistant, which makes them attractive to some.

Generally, I suspect money will continue to be pumped into this until the next short break when a few banks too big to fail need rescuing by tax payers.

 

[StefanR5R]

One more thought about what went on at this Cosmology@Home sprint:

The first 18 hours:

The occurrence of the ${URL_BASE} glitch coincides exactly with the moment from which on Gridcoin's credits/hour inclined.

I suspect what we see there is Marius trying to accommodate all those who asked for camb_legacy validation in the C@H forum,

My current understanding is that Marius actually had fixed the camb_legacy validator (or got it working well enough for the time being) at this very point in time. This means that all teams who had been crunching legacy tasks during the 4 weeks in which the validator was off got rewarded beginning Friday, 12:48 UTC. And clearly, Gridcoin naturally took away the biggest gain from this: (1.) They had a huge throughput at Cosmo before, (2.) they seem to be unable or unwilling to check on their hosts and project news in order to find out about the sudden drop in PPD when the validator was suspended, in contrast to real contributors who care enough for such things.

To summarize, Gridcoin's final FB sprint stats contain a huge amount of credits earned during the past four weeks. Other teams do not appear to have gotten that much out of it, according to my diagrams.

It is also interesting to see that our camb_boinc2docker-only PPD was quite a bit higher than Gridcoin's. And it is probable that it remained that way for the rest of the sprint.

It is obvious what would have happened if the camb_legacy validator had not been fixed before the end of the sprint. But I am left to wonder how it would have gone if the validator had been working properly during the weeks before. Could have been close between us, SUSA, and Gridcoin.

 

[Rudy Toody]

A play on words: When we sprint, we are a jogger-not!

 

[Orange Kid]

RakeSearch is this weeks sprint.

http://rake.boincfast.ru/rakesearch/ for new to project. (Like me)

 

[StefanR5R]

Gridcoin is at rank 20 of the RakeSearch marathon, IOW practically absent.
http://formula-boinc.org/projets.py?prj=221&year=2018

This should help with work availability a bit, although I nevertheless expect difficulties, from what I remember from April.

 

[ao_ika_red]

Gridcoin is at rank 20 of the RakeSearch marathon, IOW practically absent.
http://formula-boinc.org/projets.py?prj=221&year=2018

This should help with work availability a bit, although I nevertheless expect difficulties, from what I remember from April.

We did well enough without GC, SETI,USA, and CNT. I hope it will be also the case for this weekend.

 

[StefanR5R]

With SUSA/XS, [H]ard|OCP, the top two in league 3, and perhaps CNT if they are so inclined, the real stress test of RakeSearch will happen this weekend and dwarf the rather mild load that RakeSearch was given during the April event.

 

[ao_ika_red]

With SUSA/XS, [H]ard|OCP, the top two in league 3, and perhaps CNT if they are so inclined, the real stress test of RakeSearch will happen this weekend and dwarf the rather mild load that RakeSearch was given during the April event.

Remember when people started to release their bunker and jammed RakeSearch server on the last 2 days of stress test event? I imagine it will be the case of this weekend as well.

 

[DooKey]

I don't like the sound of that. I hope Rake can handle this load.

 

[Orange Kid]

I don't like the sound of that. I hope Rake can handle this load.

We shall now find out.

 

[crashtech]

I wonder how we will fare against SUSA?

 

[ao_ika_red]

I wonder how we will fare against SUSA?

No Pentathon to distract them but I'm still hopeful that this will be our next Moo!

 

[StefanR5R]

Current RakeSearch marathon standings:

3 - LAF (101 M)
4 - SUSA (77 M)
5 - P3D (68 M)
8 - OCN (43 M)
11 - RKN (28 M)
12 - CNT (27 M)
13 - The TeAm (24 M)
15 - UBT (15 M)
17 - B@Au (6 M)​

This indicates that most of the contenders are well prepared to go into this sprint.

 

[StefanR5R]

The first 18 hours:

The first hours make it look as if Russia Team, Boinc@Taiwan, and Dutch Power Cows are actively sprinting too. But I think it's simply the effect of their having wingmen who do actually participate, such that validation of RT's / BTW's/ DPC's regular output was partly shifted from Thursday to the first hours of the sprint.

 

[Orange Kid]

Day one of RakeSearch:
We are in first with SUSA coming on fast. They always seem a little slow out of the gate and then come on strong. Will they continue the steady climb, will they have enough left on that last furlong? Will be an interesting one !!! There is sure to be some shuffling in the lower ranks before this ones over too. Wish I could send some of these cooler than average temps to you all in the heat.

1 TeAm AnandTech ...............….25 ...…...….2,280,803
2 SETI.USA ...........................…..18 ......…….2,211,177
3 Rechenkraft.net ........................15 ............….703,838
4 Russia Team .....................…….12 ............….696,769
5 Czech National Team .........…….10 ...............572,778
6 Boinc@Taiwan .....................….....8 .........…...470,370
7 Dutch Power Cows ..................…..6 ...............388,753
8 Planet 3DNow! ..............................4 ...............377,258
9 Overclock.net ...........................….2 .........…...304,676
10 BOINC@AUSTRALIA .........…….1 ...............286,747

FreeDC stats
emoga ..399,488
crashtech ..387,607
Howdy2u2 ..195,821
biodoc 158,363
10esseeTony ..111,888
Orange Kid ..77,520
Ken_g6 ..51,208
Mark ..33,627
xii5ku ..33,074
Rudy Toody ..26,117
GLeeM ..15,464
zzuupp ..2,028
Fardringle ..41

​

 

[Bradtech519]

I've joined the race. I used the AVX2 program. it's showing up as Local: RakeSearch (Opti v1.0) 7.02. Seems like the tasks are going to take longer. I have four tasks that will complete in 6 hours 13 minutes. Seems like the default I got was sooner. Hopefully the AVX2 work units give more points?

 

[Bradtech519]

nevermind.. I think the Remaining Time is way off for Boinc. I'm already at 10%.

 

[ao_ika_red]

1 TeAm AnandTech ...............….25 ...…...….2,280,803
2 SETI.USA ...........................…..18 ......…….2,211,177

I hope this position will stay the same until the sprint has finished.

edit: Nope, it's not.

 

[Bradtech519]

My tasks should start coming in. Hopefully will help.

I hope this position will stay the same until the sprint has finished.

edit: Nope, it's not.