9th Annual BOINC Pentathlon 2018 Strategy & Preparation

pututu

pututu

[H]ard DC'er of the Year 2021
Joined
Dec 27, 2015
Messages
2,880
Believe it or not, the 9th Annual BOINC Pentathlon 2018 is just 45 days away. It normally starts on May 5th (penta-month and penta-day) of every year and lasts for 2 weeks. This is hosted by Seti Germany (SG). The countdown timer can be found in SG website. There are five events (pentathlon) or projects that will be announced (normally 3 or 5 days ahead) and bunkering is allowed, so you should at least be familiar with various bunkering techniques. My guess is that the rules are likely to be the same as last year. See 2017 post here and here for more details about last year event.

Team participation is not automatic. I believe Gilthanis will register our team. No need for individual registration as this is a team event.

Between now and when SG officially announces this event, there are couple of things that we can do. Here are my thoughts and please chime in for the benefit of other [H]ordes.
- Think of what project we want to vote on. Last year projects are not eligible for this year. So no Cosmology, YoYo, WCG Open Zika (other WCG projects may still be possible for this year even), Einstein and LHC. For some unknown reason, every year, there are a few lousy projects that got chosen due to its popularity (?)
- Familiar with DC projects via FB participation.
- Oh, if you decide to upgrade your rigs, now is the time do to it. That will be very helpful to the team.

BTW, Gridcoin does participate this but they never got the top spot since this requires them to run some non-whitelist project, frequent switching of projects and perhaps they don't know how to bunker :D.

It will be an exciting event this year as I do see a renewed interest in DC by [H]orde members.
 
Me thinking two get iTX boards with 8700(K) dedicated for CPU crunching; alternating with GPUs usage to keep power cost under control; would those help ?
 
Other than the ARM and raspberry pi processors, for me I still think the used xeon high count cores are the way to go if you can find them at reasonable price for CPU crunching, in terms of power efficiency and PPD. In the future, I wish more DC projects support GPU (with optimized applications), so I think having more than 2 GPU slots and running these cards with power limit is the way to go for efficiency and heat dissipation over the long term. I never own an iTX board but I think it can only accommodate one slot. GPU prices are crazy now even though there is some price drop recently. Just my one cent.
 
I'm getting a bit board with all the prime-math stuff; and on BioHealth there is primary GPUGrid and WCG; even GPUgrid right now has more work for CPU ... funny enough. and the 8700K seems has a good single-core performance in the overall package. Helping for WCG is the hope. and of course other single-process stuff too. ITX just for place saying; no GPU planned on those board. Would be only got CPU crunching. My two GPUs are still served well with my i7-2600S ...
 
We are effectively 1 month away from this. I plan on creating a public thread on this later this month. I want this on everyone's radar as it is THE BOINC CHALLENGE of the year. This is where people test their metal and stand up to the biggest of the big . Yes it typically hits everyone in the nuts with Summer weather, but luckily it is only for a few weeks. Timing is important. Tweaking your PC in advance is important. Being ready for quick movement is important. And I cannot stress enough how critical bunkering is on this challenge. Every team on the board will be doing it and doing it in mass. If you need to learn and tweak a favorite way of doing bunkering, use the FB sprints leading up to the Pentathlon to do it. Again, bunkering tactics and efficiency tends to be the biggest strength in the Pentathlon. I also recommend having a few different VM's set up as fall back options and also highly recommend using Linux when possible as many projects score much better with it. Even running a Linux VM will many times out produce Windows on bare metal. I'm thinking of setting up a few soon, but haven't fully decided. I'm also thinking of getting a few quads sitting cold updated (though they are Win10) to fire up for those two weeks. I can probably bring 3 or 4 quads online right now and may have another 2-4 quads to bring too. But That will probably be my limit on adding new gear unless I'm lucky enough to get GPU upgrades or stumble into a nice CPU/box upgrade. I would love a dual socket LGA2011v1 or v2 motherboard falling my way...lol. As I have a low power chip with 8c/16t CPU that I would find a mate for. but I don't have any serious need for 2011 or newer for any other reason at this point. I have plenty of DDR3 RAM, so any upgrades needing DDR4 for me would be significant $$ investment that I cannot justify for the hobby.
 
Talking about hardware upgrade, all in all, I paid about $70 to upgrade my 2683 to 4699. Sold the former recently. Didn't break even.:(

Now looking for GPU bargain but will have to wait since the GPU is still hovering close to MRSP for some cards but the new gen card hopefully will be announced soon...
 
I also do not have a lot of GPU power. So, the only cards I can toss into old rigs would most likely be the lowest of the low in the 8400 GS level. Though better than many CPU cores, they typically aren't worth the power draw. If anyone has old GPU's that are collecting dust, please consider putting them to work where possible for these 2 weeks of competition. Our team is typically GPU lite compared to our competitors. People who were in Pentathlon's in the past may want to consider strategy and also decide if we think it best to try and finished with an overall higher score or try lumping all of our eggs into a single basket to try and take a single event. More of that discussion will occur as things progress. But, I just want people thinking about it. For the newcomers coming in, it is unlikely we will take a top 3 spot overall unless we can find a hand full of fastgeeks and/or a ton of miners with setups more fitting to DC liking.
 
I think we should start thinking about the project nomination (5 in total). I hope that after the event, for transparency purpose, they should publish the list of projects submitted by each participating teams. I didn't see a list of this in 2016 and 2017 event. Sometimes for unknown reason a lousy project got selected (lousy means has little work, server unable to cope, etc).
 
I believe they said they were going to publish that info but don't recall ever seeing it either. But my mind could be murky on the details. They may change the rules up again. So, we will need to wait until there are more details.

Edit: Nevermind, they have some listed already. Public post under construction
 
Looking at the projects on the list so far.... I'm gonna find it hard to get many people on our team interested as it is Bio/Medical light and very Mathematics heavy this year.

Quorum CPU GPU
Quorum 1 NFS@Home
NumberFields@home
ODLK*
Rosetta@home
SRBase
YAFU ...
Quorum 2 Asteroids@home**
Citizen Science Grid
RakeSearch
Universe@Home
Amicable Numbers
PrimeGrid
SETI@home

This list could still change but isn't very impressive for this large of a challenge. I was hoping with the way WCG was restricted last year to just one sub project that they would either do that again or allow the full project to qualify this year. I guess not.

Personally, I think CSG is a bad choice because the last time it was chosen, they had trouble and took a while to get things resolved. Since that is right at the end of the semester, Travis may not be available for full support. I'm not a fan of ODLK nor am I convinced they could handle the requirements. From the sounds of it, they may utilize both projects as one competition which will be terrible for users but definitely a strategy game changer. YAFU will be a problem at least at the beginning unless YoYo took last years scenario and planned for it. But since it is on test server software.... I'm guessing more headaches. Not to mention trying to explain to everyone how to limit cores with app_configs and such. SETI would work but you will have people bitching about work unit validation and of course whether it rolls on the weekly server maintenance or not. The rest would probably suffice in some manner, but even with the biggest of projects Murphy's law typically kicks in. lol
 
Last edited:
After the FB sprint is over, I would like somebody or everybody to chip in on some numbers for the projects above. I think this year we should have a more organized strategy. For example, how long are the deadlines for each of the above projects' applications. How long do the work units typically run? Do the work units validate quickly? Are there quicker way to run the applications? For example.... PrimeGrid you can set up an app_config to run the work units multi-threaded instead of one per core. This is handy if you have slower CPU's that don't have better AVX2 extensions (*cough* most AMD *cough*). How many cores dedicated to a single work unit provides the best bang for your buck? I know YAFU is a crapshoot on this one. However, years ago I personally had better luck running 4 cores per work unit but that was before YoYo created the 4t, 8t, and 16t work units. We need to know if projects will prefer HT on or off and whether that still applies if running a multi-threaded app_config file or not. We have less than a month to iron some of this out so would be best if it was a joint effort. We need to know how to direct everyone for when new members hop on board. Also, knowing if a project has an optimized app that truly does improve things would be good too.

As of right now:
Rakesearch has optimized apps - I will post the link later from home
PrimeGrid has the ability to run multi-threaded - I will post a large app_config later when I have time to dig it out.
SETI - sloooooooooow validation times for some work units. Astropulse are known to score much better but are hard to come by most of the time. Also has a maintenance day every Tuesday. SETI has some optimized apps but some work units may benefit more than others.
Android applications - Rosetta, SETI, Asteroids typically have work but YAFU rarely does.

Also, when it comes to PCIe data lanes and throughput, Linux seems to be much faster from the reports I have seen. So, if you have multiple cards fighting for bandwidth, it may be best to run Linux. Linux also scores better at many projects. We should get more difinitive numbers for each project. I for one wouldn't mind setting up some Linux VM's but will not be installing it on bare metal.

Once we have a collection of data, we can possibly even have pututu edit the first post or we can designate a post to update with all the details in one place. If we want it public during the event or leading up, I can put it in the Pentathlon thread. But I didn't know how much of that info we wanted out there before hand. We can always put it in the open forum after the event too.
 
Last edited:
Do you know if the project suggestion list is final?

upload_2018-4-6_12-50-38.png

For Bio/medical, rosetta would be a good choice. The rest I'm impartial but would like others to provide feedback.

I was thinking about having 5 + 1 threads in our private forum. The 5 threads will each represent the project selected and detail about optimized setup, performance, issues, problems, etc specific to that project. In this way, it will be easier to search for useful information to run this specific project, if needed, These 5 threads should be limited in posting as not too long. The 1 thread can be strategy discussion such as this one or a new one.

Recruitment should be in our main forum so everyone can see. Nothing ot hide here other than bragging, complaining, swearing (might get deleted if doesn't meet the forum guidelines), cursing, praising, etc.
 
Last edited:
I am guessing so as registration is live now.

Edit... looking at the selections for registration..there are a few more that have been added. I updated the list
 
Last edited:
FYI, there will be one FB sprint from May 10 - 13 during the Pentathlon. If there is a resource conflict (either CPU or GPU or both), my take is to forgo the FB sprint if we are in a very tight spot (I'm pretty sure we will be considering we are up against the big boys). Just wanted to share my opinion early.
 
An interesting strategy used by ChristianVirtual is to rent Amazon EC2 (CPU) and maybe GPU on Google Cloud. I think we saw huge spike in output from some teams in last year Pentathlon which we think may originated from the use of data centers/cloud computing. Now the question is should we employ this tactic? Seems unfair and defeats the purpose of the BOINC main purpose of "volunteer computing" but [... fill in the blanks... ]

Perhaps this is more of a philosophical discussion.
 
EC2 instance usage certainly brings up an ethical talking point, no doubt about that....but at the end of the day, this is all still for good because science can use these tactics ;)
 
It was an ethical question like 10 years ago. Nowadays it is looked at no different than using you jobs systems. Might even looked better upon because it is guaranteed to have permission to use. Everyone has access to doing it. I say, have at it. Oh and I got this today.

hi [H]ard|OCP,

it's that time again, the BOINC Pentathlon is just a few weeks away. For
the ninth time, SETI.Germany invites the passionate crunchers from all
over the world to compete in this team competition from 05 May to 19 May
in five disciplines at five projects.

The teams may sign up and vote for their favorite projects from a
pre-selection using the registration form at
https://www.seti-germany.de/boinc_pentathlon/anmeldung.php until 27
April. Late registration until 02 May is possible, but without the
opportunity to vote for projects.

For more information about the BOINC Pentathlon, please visit the
Pentathlon pages: https://www.seti-germany.de/boinc_pentathlon/

It will be exciting again at the BOINC Pentathlon! Looking forward to
seeing you again there! :)

Regards
pschoefer
SETI.Germany
 
If we can recruit some rich "volunteers" or donors and rent the EC2 for us, we might have a shot of being on top three :D Maybe that's a new strategy [..sarcasm...]

Once you buy the plan, does anyone know how easy to setup BOINC projects (CPU only, right)?

ChristianVirtual, if you can help to write a simple guide, that will be useful in case we need to use it during the event. Maybe we should do some practice run first.
 
Using EC2 I don't see unethical more then having multiple GPUs or CPUs boxes at my home or rented root server in a DC. Its just a question how I spend my (direct or indirect) resources to get something out.

The risk with EC2 instances is that, when you go like me with spot instances, you risk to loose the system during the run. Thats why for a sprint its controllable; fire the cloned instances up and keep them running as long then credit card allow.

For a bunker concept I would not go for it as you at risk not only loosing the actual work in progress but also the bunker. That would be waste. Sure, you can have a storage attached in Amazon world to prevent; but getting more costly. So overall the individual question one need to answer: do I have the cost under control.

To visualize what I mean: the current cost graph for this four instances over the last two days. I set a cost limit by 1.5 USD/hour for each instance.

Screen Shot 2018-04-08 at 6.14.57.png

Yes, I had fun and from time to time its worth doing it. Would I do it on this scale for longer like Pentathlon ? Maybe not. Some other sprints ? Maybe yes.
 
ChristianVirtual said:
I set a cost limit by 1.5 USD/hour for each instance.
Click to expand...
I salute you for doing this with great passion
upload_2018-4-7_14-32-50.png


If I understand it correctly, 1 instance that you are currently renting has 72 CPU threads. A full day for each instance will cost $36 based on current spot price.

May find this useful if we are short of a few points to take a decent spot during the last few hours of the race but certainly I wouldn't spend days and weeks running BOINC with EC2. Too expensive for my taste.;)
 
The real question is how much "free" can a user get even if on a trial basis and how many people can we get to set up an account if there is an easy "how to" with pictures.........
 
Last edited:
I don't see any ethical issue running EC2 instances. It's not as if you're getting rewarded for doing it. There's no difference in my mind between spending money on these or your own hardware/electricity to crunch science. It may be a sensible option, for short bursts. in place of capital expenditure.

The problem may come when, in a blood rush to win a challenge, teams start firing up instances they can ill afford. Even then, the problem will only manifest itself when the wife sees your credit card statement or gets a slab of cheddar cheese for her birthday.

Gilthanis said:
After the FB sprint is over, I would like somebody or everybody to chip in on some numbers for the projects above. I think this year we should have a more organized strategy. For example, how long are the deadlines for each of the above projects' applications. How long do the work units typically run? Do the work units validate quickly? Are there quicker way to run the applications? For example.... PrimeGrid you can set up an app_config to run the work units multi-threaded instead of one per core. This is handy if you have slower CPU's that don't have better AVX2 extensions (*cough* most AMD *cough*). How many cores dedicated to a single work unit provides the best bang for your buck? I know YAFU is a crapshoot on this one. However, years ago I personally had better luck running 4 cores per work unit but that was before YoYo created the 4t, 8t, and 16t work units. We need to know if projects will prefer HT on or off and whether that still applies if running a multi-threaded app_config file or not. We have less than a month to iron some of this out so would be best if it was a joint effort. We need to know how to direct everyone for when new members hop on board. Also, knowing if a project has an optimized app that truly does improve things would be good too.

As of right now:
Rakesearch has optimized apps - I will post the link later from home
PrimeGrid has the ability to run multi-threaded - I will post a large app_config later when I have time to dig it out.
SETI - sloooooooooow validation times for some work units. Astropulse are known to score much better but are hard to come by most of the time. Also has a maintenance day every Tuesday. SETI has some optimized apps but some work units may benefit more than others.
Android applications - Rosetta, SETI, Asteroids typically have work but YAFU rarely does.

Also, when it comes to PCIe data lanes and throughput, Linux seems to be much faster from the reports I have seen. So, if you have multiple cards fighting for bandwidth, it may be best to run Linux. Linux also scores better at many projects. We should get more difinitive numbers for each project. I for one wouldn't mind setting up some Linux VM's but will not be installing it on bare metal.

Once we have a collection of data, we can possibly even have pututu edit the first post or we can designate a post to update with all the details in one place. If we want it public during the event or leading up, I can put it in the Pentathlon thread. But I didn't know how much of that info we wanted out there before hand. We can always put it in the open forum after the event too.
Click to expand...

I've been capturing some data and thoughts since I was trying to decide what to bunker for the new year. The ppd/runtimes are based upon all cores of a threadripper with HT on or a 1080ti running at 55% power usage.

Boinc Data Capture 08-04-18.png

The data is only as good as the last time I ran the project and so may be out of date. It also doesn't cover everything you mentioned such HT on/off or time to validation but hopefully it's useful as a starter for ten to correct as necessary and build upon.
 
phoenicis said:
It may be a sensible option, for short bursts. in place of capital expenditure.

... Even then, the problem will only manifest itself when the wife sees your credit card statement or gets a slab of cheddar cheese for her birthday.
Click to expand...
Exactly my thoughts on the first part on economics; as for the second one: my wife would slap the slab of cheese back to me
 
for the guys running Rosetta@home right now... can you take a moment to play with some settings. In the Rosetta@home website, change your preferences to dictate how long you want work units to run. I believe the default is 8 hours. You can bump them all the way up to 1 day. You can change this all the way to as low as 1 hour. I am essentially wanting to see if the scoring per hour is roughly the same or not. From what I can muster, the work units running short or longer are essentially the same but the longer you let them run the better the science is. However, for the sake of the Pentathlon, there may come a period where everyone needs shorter work units to guarantee they get turned in. Also, we need to know if there is a download limit for Rosetta. The last time we verified pututu confirmed a limit of 100 per host. This means if we need to bunker this project, you may want to up the processing time to something high to begin with but maybe something low after dropping the bunkers.

rosetta.PNG

What really sucks is that the most popular CPU project voted for will get the shortest amount of time to be ran. So, here is to hoping that Rosetta is picked by SETI.Germany as the Marathon or it will most likely be a 3 day event....lol

And for reference, this is exactly what happened 2 years ago. It only ran for 3 days.
 
Last edited:
This is probably going to get shot down.

Recruiting more volunteers in the Pentathlon, in my opinion is going to be key to our good finishing. Wondering if we can make [H] readers/members aware of the upcoming Pentathlon and encourage them to join us. Here are my ideas/suggestions but I have a feeling that it will be denied. At least I tried...

Is anyone considering Amazon EC2?


upload_2018-4-20_21-12-22.png


or

upload_2018-4-20_21-13-17.png


or

upload_2018-4-20_21-18-44.png
 
That last one would probably be the most likely one that Kyle would agree too, but even then I doubt he will.
 
In about 22 hours or more, the first marathon project will be announced. Project can be announced at these times: 00:00 UTC, 06:00 UTC, 12:00 UTC, 18:00 UTC.

Current UTC date and time can be found here. Remember first project will be announce sometime after Apr 30 00:00 UTC time.
 
I noticed this over at TAAT's page

If my math is right, the Marathon announcement = 5 days before the start will be at the earliest on
Monday 10:00 AEST, 07:00 WIB, 02:00 CEST, 01:00 BST, 00:00 UTC,
Sunday 20:00 EDT, 19:00 CDT, 18:00 MDT, 17:00 PDT, 14:00 HST.

(That's 4 hours after the Formula Boinc sprint ends.)

However, throughout the Pentathlon, project announcements will be randomly deferred by up to 18 hours. This is so that contestants at all parts of the world will be kept on their toes and robbed of their sleep... ;-) I am not sure if they'll opt to defer the marathon project announcement too.

Edit April 29,
according to pschoefer at the shoutbox of the Pentathlon site, the marathon project announcement will be exactly at the time I noted above, IOW it won't be deferred randomly.
Click to expand...
 
Pretty soon we will know the CPU marathon project. Any idea how to get this event known to larger audiences in [H] forum other than the regular folks here?

The marathon project is ideal for those who just want to help [H]. Just need to set this up once and let it run till the end of the pentathlon. No need to switch project frequently, bunkering, disable network/block host, etc.

I'll finish all the projects that I'm currently running today and let the queue be empty. I just like to start this fresh and clean.

Thoughts?
 
Same way I recruit people to the Christmas Race and the Formula BOINC Sprints.

I'll make a post in General Mayhem and see where it goes.
 
You must log in or register to reply here.
Back
Top