7th Annual BOINC Pentathlon (2016)

Gilthanis

[H]ard|DCer of the Year - 2014
Joined
Jan 29, 2006
Messages
8,759
-
boinc_pentathlon2016.png
/
7th Annual BOINC Pentathlon - BOINC Pentathlon > Welcome | seti-germany.de

Hosted by Seti-Germany
06/05/2016, 0:00 UTC to 06/19/2016, 0:00 UTC
5 disciplines, 1 winner
Place Holder for official projects once they are announced






I will be signing us up for this but wanted to get some feedback from the team. Last years projects are not eligible this year. We get to vote for 4 projects. The projects with the most votes will be added. The following projects are what we have to choose from:

Asteroids@home
ATLAS@home
Citizen Science Grid
Collatz Conjecture
Cosmology@home
DENNIS@home
distributedDataMining - Removed due to my suggestion and past experience with this project and challenges
Enigma@home
GPUGrid
MilkyWay@home
Moo! Wrapper
NFS@Home
NumberFields@home
POEM@home
PrimeGrid
Rosetta@home
SAT@home
SRBase
theSkyNet POGS
Universe@home
VGTU@home
vlhc@home
YAFU

Updated details on project selections for us to vote for 4/7/16

The BOINC Pentathlon consists of 5 disciplines:

  1. Marathon (CPU, 14 days)
  2. Sprint (CPU or GPU, 3 days)
  3. City Run (CPU, 5 days)
  4. Cross Country (GPU or CPU, 5 days)
  5. Swimming (CPU, 7 days)
Each team may enter 4 project suggestions while signing up. The projects are assigned to the disciplines as follows:

  1. Marathon
    • The project is set by the organizers.
  2. Sprint
    • The project with the most votes that provides workunits with a quorum of 1. Both CPU and GPU projects are possible.
  3. City Run
    • The CPU-only project (i.e. the project does not provide any GPU application) with the most votes that is not already chosen for the sprint.
  4. Cross Country

    • The GPU project with the most votes. If this project has been chosen for the sprint, the project (CPU or GPU) with the next most votes that has not been chosen for another discipline is chosen instead.
  5. Swimming
    • The CPU-only project (i.e. the project does not provide any GPU application) with the most votes that is not already chosen for another discipline.
CPU-only GPU support
Quorum 1

ATLAS@Home
Cosmology@Home
Enigma@Home
NFS@Home
NumberFields@home
Rosetta@home
SRBase
VGTU@Home
vLHCathome
YAFU

Collatz Conjecture
GPUGRID
Moo! Wrapper
POEM@HOME

Quorum 2
Citizen Science Grid
DENIS@Home
SAT@home
theSkyNet POGS
Universe@home

Asteroids@home
MilkyWay@home
PrimeGrid
SETI@home

Projects are not eligible if they

  • were chosen in last year's Pentathlon (Einstein@Home, LHC@home Sixtrack, malariacontrol.net, World Community Grid, yoyo@home)
  • can not provide a sufficient number of workunits
  • support just a few operating systems
  • provide non-CPU-intensive (NCI) applications
  • use buggy or test-only applications
  • do not allow the creation of new accounts
  • do not support WebRPC
To keep it as transparent as possible, we will publish the suggestions of every team when announcing the final project.
 
Last edited:
For those not familiar:

The following projects have virtualbox work:
ATLAS@home
Cosmology
vlhc@home

The following projects have shown to be poor candidates in the past for challenges:
distributeddatamining

The following projects go long spells at times without work:
theSkyNet POGS
Universe@home
Sometimes Citizen Science Grid

The following projects have multi-threaded apps:
YAFU
Milkyway@home
Collatz Conjecture

The following projects have GPU work units:
Asteroids@home
Collatz Conjecture
GPUGrid
Milkyway@home
Moo! Wrapper
PrimeGrid

The following projects have Android ARM work units:
Asteroids@home
Citizen Science Grid
Collatz Conjecture
Enigma
Milkyway@home (if using nativeBOINC only)
Moo! Wrapper
Rosetta
theSkyNet POGS
YAFU (sometimes but not very often)

The following projects may require 1GB+ RAM per work unit:

ATLAS
Cosmology
NFS@home
vLHC@home
 
Last edited:
My top four from this list are:
Asteroids@home
Milkyway@home
PrimeGrid
Rosetta

My reasonings are:
1. Asteroids and Milkyway both support a wide variety of hardware. Both are stable while providing typically plenty of work. Milkyway requires double precision video cards for GPU work so AMD is king. Milkyway also has multi-threaded apps so people with large boxes might prefer this kind of work. Last I knew, Asteroids only supported nVidia GPU's. However, my backup choice is not Asteroids would be GPUGrid.
2. 3 of these projects support GPU's which our team is finally starting to invest in. All four support CPU's.
3. All four projects are typically well ran, active admins, and can supply work.
4. PrimeGrid has many challenges through the year so if people were sick of running it, my backup would be Collatz Conjecture.
5. Rosetta is still a bio/med project that many on our team have supported. It is a solid and well ran CPU project.
6. I feel that out of all of the projects listed, these will require the least amount of Education overhead, least general user headaches, most reliable, and most likely to have enough work available.
 
Last edited:
So, with the new information provided, my suggestions would be:

Sprint - GPUGrid
City Run - Numberfields
Cross Country - POEM
Swimming - Rosetta

I figure 3 out of those four are BIO/Medical projects. 2 of the 3 bio/medical projects have CPU work and 2 of the 3 have GPU work. Numberfields is CPU only but ran pretty solid for quite some time. Anybody else got ideas or suggestions. I would prefer to have the teams opinion on this before I cast the vote at signup.
 
Please give me another day to think about this, if you can.
 
We have some time before the voting ends, so you certainly can think about it.
 
My latest build with one of Patriots E5 CPUs happens to have an R9 Nano installed. I would love to get Milkyway running on it but have a snowballs chance in Hell of getting AMD drivers installed on Ubuntu 14.04.
 
You could slap Windows on a spare drive for the challenge.... :D
 
Actually, Windows will give you 30 days without activating with a key..... lol ;)
 
Ok, I found a 120GB SSD that had Windows 8.1 on it. Fired it up and it says it's activated. Not sure how that happened. Maybe it didn't phone home yet. Because of my data cap I can't do any updates until midnight. I'll try updating it then and hopefully eventually get Win10 installed. If all goes well I'll run Milkyway on the R9 Nano and WCG on the CPU. Why isn't WCG part of this upcoming challenge? I'm currently doing about 110K a day on WCG. And I'm not willing to switch from WCG.
 
Because WCG was part of it last year. One of their rules is that the project couldn't have been part of the previous year's Pentathlon.

Projects are not eligible if they

  • were chosen in last year's Pentathlon (Einstein@Home, LHC@home Sixtrack, malariacontrol.net, World Community Grid, yoyo@home)
  • can not provide a sufficient number of workunits
  • support just a few operating systems
  • provide non-CPU-intensive (NCI) applications
  • use buggy or test-only applications
  • do not allow the creation of new accounts
  • do not support WebRPC
 
Bummer. I really don't want to switch off WCG but might consider it if Milkyway would run decent on my collection of CPUs. On CPUs how does Milkyway PPD compare to WCG PPD?
 
Been a long time since I ran CPU on MW, but WCG is well known for being a low scoring project.
 
Would Rosetta be a better choice for this challenge on the CPUs? Been a while since I've run it.
 
Rosetta is my choice for this challenge on CPU's. Just keep in mind their work units can hover around 512MB each at times. But we have to vote for projects. So, it is mostly up to what everyone else chimes in for preference... and even then... our team only gets the one vote and when all votes are in..that dictates what projects are actually associated. And they announce them officially slightly before they begin. So, it is a surprise to everyone on which will be.
 
So far I favor Rosetta and GPUGRID as I've crunched those in the past. Milkyway would be fine as well.
 
Sprint - POEM
City Run - SRBase
Cross Country - Milkyway
Swimming - DENIS

This gives us projects in each of the following disciplines: Bio/Med Science (2), Mathematics (1), and Physical Science (1). My only apprehension is with DENIS, as they may not have work. My second choice would be Rosetta, which seems to always have work.

Is bunkering allowed for this? I won't do it myself, if so, but I like to know what we are up against.

Any chance some of the Folders on Team [H] might move over to BOINC for a month? We need some horsepower for this, or we will get smoked. We just don't have all that many on the team who are crunching BOINC projects anymore, unfortunately.
 
I meant to write earlier that I picked POEM over GPUGrid for the Sprint, because the Long WU's at GPUGrid can take as long as 16 hours, or so, on my 780Ti. For a three day challenge, there wouldn't be much work completed. Last I checked, GPUGrid often has no Short WU's available, so everyone would be running the long WU's. In contrast, POEM WU's are much shorter (~.5 hour per WU on my 7870/280X GPUs), so are more appropriate for a "Sprint."

GPUGrid also runs only on Nvidia GPU's, while POEM can run on both AMD and Nvidia GPU's, as well as on CPU's.
 
Bunkering is allowed and yes pretty much every team does it. It makes a HUGE difference in this event. Last year, bunkering was the sole reason we didn't place in the top 3 of any of the categories. Otherwise we would have had a pretty good chance at the MalariaControl one. So regardless of standpoint on bunkering, you may want to consider it for this challenge. Some projects like Rosetta that have a very strong back bone and an unlimited supply of work are typically unphased by the tactic.
 
ATLAS@home is asking for teams to vote for them. I don't see our team getting excited as it is one of the virtualbox projects. However, I don't see a lot of teams excited for that, so who knows how much competition there will be on that if it happens.
 
I like ATLAS, but the problem, as you know, is most machines can't run more than a few ATLAS WU's at a time due to high memory and I/O usage. Therefore, I don't think it is a good project for this type of challenge.
 
So with the little feedback from the team given. RFGuy_KCCO and I talked it over and I let him pick the final listing.

Sprint - POEM
City Run - Numberfields
Cross Country - GPUGrid
Swimming - Rosetta

Those are the projects we are hoping to run. Granted our choices are only votes, but we felt those would probably mesh the best.
 
As posted here: 7th BOINC Pentathlon Delayed

Dear fellow crunchers,

bad news today. As you may already have noticed, seti-germany.de and the Pentathlon pages have been down since Saturday. The server was hacked and we do not know yet when everything will be back online. Of course, we do not want to bring it back up just to see it being hacked again during the Pentathlon. Additional security measures are being implemented right now.

I really feel bad to have to say this, but we will have to postpone this year's Pentathlon to make sure that we can provide the usual service (stats, daily bulletin, project announcements) and all teams have the opportunity to sign up. We can not give a new date right now, but we hope that we can run it still in May or early June, as we definitely want to avoid the hot summer months.

If you registered your team before last Friday, all information is in our latest backup and will be restored. If you wanted to participate but could not sign up yet, don't worry; we'll make sure that you can still sign up once the Pentathlon pages are back online (of course, you can also submit your project suggestions via email if you like).

We will post a proper announcement as soon as we have a new date. Really sorry for the chaos this year.

Regards,
pschoefer
 
With that news it looks like I won't be able to participate at all this year. I go to my summer home in CO around June 1. All my servers in TX will be shut down for the summer.
 
update

SETI.Germany Forum - BOINC Pentathlon 2016 rescheduled to 05 June 2016

BOINC Pentathlon 2016 rescheduled to 05 June 2016
von pschoefer
Veröffentlicht: 01.05.2016 20:00
Dear fellow crunchers,

unfortunately, the BOINC Pentathlon 2016 has to be postponed for a month after a hacker had caused havoc on the server and the SETI.Germany website had been offline for a week. Therefore, the new schedule is:

Start: 05 June 2016, 0:00 UTC
End: 19 June 2016, 0:00 UTC

The registration deadline for all teams that want to participate has been adjusted accordingly: Project suggestions are now accepted until 28 May. The registration form is working again. All teams which had already signed up should still be on the list of registered teams.

There will be additional efforts during the next weeks to recover the SETI.Germany website completely, thorough functionality testing of the Pentathlon pages will also be performed again. The postponement makes sure that there is ample time for our efforts and for those teams which could not sign up in time due to the server downtime to sign up.

We apologize for this incident and the postponement, but are still looking forward to run a thrilling BOINC Pentathlon with you a month later.
 
Kind of quiet here. Anyone signing up for this Pentathlon?

Just saw this posted at BOINC Pentathlon site. Citizen Science Grid is the marathon project. Any useful preparation tips?
 
Good to know about how credit is granted. I'll be running a few hours before the official start time. Just warming up:)

"All credits granted between
05 Jun 2016, 0.00 (UTC) and
19 Jun 2016, 0.00 (UTC) will be taken into the BOINC Pentathlon validation. Please note that pending credits cannot be taken into account! "
 
CSG typically has plenty of SubsetSum work units which also works on Android devices pre v. 5. DNA runs out of work often. It's upload files last I knew could get up to 20MB if that matters for anyones data limits. Wildlife hasn't had work in years. The Twitter stuff doesn't earn points (or at least last I looked at it). Watching the videos doesn't earn point either.

My only issue here is that with the semester being over and summer term in session, Travis may not be accessible. I am hoping he is aware of the load and demand that is going to be put on the project. Being the 14 day portion, it could make for a very rough challenge.

Keep in mind that if you have work units pending beforehand, that those will validate during the challenge. Also, bunkering is pretty much necessary if you want any shot at competing as this is the favored tactic of all the teams.... For those not knowing what bunkering is, please have a read: Bunkering - Why you should or should not do it

Edit: Team page link added to first post.
 
Last edited:
All of the work units have already been hoovered up. I did shoot Travis a PM in hopes he can address things in time.

Edit: I've got this gut feeling that this year will be less about how much you are capable of crunching and more about tactics. Tactics including blocking others from having work to process. So, virtual machines can come into play to gobble up work even if they aren't intended to be crunched during the challenge.... A few years ago some teams used the bunkering technique at POEM to prevent people from having anything to crunch. I'm not a fan of that, but it certainly is a possibility.
 
Last edited:
I should have checked my notes first...

Citizen Science Grid allows 4 work units per core and only 10 will download at a time. Total machine limit is 256 work units. - 5/29/15 via PM from admin

That will severely cut down on bunkering unless he has made changes since then. However, the round about way of still doing it is to create a bunch of VM's so that they can all pull the max, suspend them, and then resume them as you run low on work from other VM's. Then when time comes...set them all to No new work and then turn on networking to report all of them one at a time.
 
Last edited:
Just to clarify on the limit: is this based on CPU cores or threads? If cores, AMD has the upper hand (but slower crunching?) than Intel in terms of the number of WUs that can be downloaded.
 
The clarify, they mean CPU threads as they show up as "CPU's" just as cores do within the BOINC client.

So, a 48 core AMD rig can only max out at 192 work units at any given time regardless of cache due to the 4 work units per core rule. You wuld need more tha 64 threads in a box before you would hit the 256 limit. Not a lot of boxes out there above 64 threads and that is easy to get around using a second client or a VM.


And this is assuming you aren't making your client lie about how many core/threads it really has. Which is a simple line in your cc_config file........ I've used test VM's that claimed they had 80 cores before but only had 1 CPU core assigned to the VM. It is easy to do and is one way people bipass the per core rule. However, the total machine rule is a hard limit that you can't overcome without virtual machines.
 
Some other tips
1. SubsetSum is fixed credit
2. SubsetSum (IIRC) runs faster in Linux and thus scores higher PPD
3. DNA is a longer running task than SubsetSum and scores higher but can vary in awards
4. DNA under Linux uses more RAM
5. DNA scores better under Windows and can be confirmed at WUProp
 
Hmm trying to get some WUs but to no avail. "0" unsent task

upload_2016-5-31_20-16-54.png
 
Boy do I hope Poem gets chosen... brilong is putting up like 5.7 million in RAC right now

I think Travis is clearing out the previous run to start with a new batch of work. At least that is what people are alluding to currently...
 
The 5 day city run has been announced - theSkyNet POGS
First post has been updated

POGS is a solid running project. It has been known to run out of work regularly so crossing our fingers they have prepared accordingly.

This project supports Android devices as well. Both PIE and non-PIE capable. This project does not have GPU work units. Server status currently only shows 139 work units to send....
 
Last edited:
Back
Top