9th Annual BOINC Pentathlon 2018 Strategy & Preparation

ChristianVirtual said:
Asteroid get very red on our six...
Click to expand...

And this is what I was worried about as I removed each box from this project. Kept telling myself that I have a huge pile of WUs awaiting validation, and that any big bunker drops will help people like me, too... but still makes me really nervous.
 
So come 20 hour from now, should I move my GPUs over to Asteroids for the last day? Of course keep the CPUs crunching towards Numberfields for that last day.
 
yeah the last day you can move the GPUs to Asteroids.
 
Watch our back!

upload_2018-5-16_21-7-50.png
 
Now truth is, "Dream Killer" has damn near killed me. Tried four OS'es before installing 2016 and that took for...ev...er. Not sure why; but suspect my mistake was mounting a network ISO as the virtual IPMI optical drive as there seemed to be some sort of network congestion going on. Should've just burned a damn DVD and been done with it. On top of that, reboots take well over 5 minutes, which is just a little tedious. Oh. And it's loud. My hearing sucks... so if I say it's loud... it is. :p

Anyway. While Windows sees all 192 threads, and BOINC has been told to use them, it would appear to be using just slightly more than 128. So am installing HyperV now. Will let 128 threads run natively and the other 64 via the VM. Hopefully it'll work as I'm pretty f'ing sick of this thing. It's 0200 and I'm not home. This better be worth the trouble!
 
Who need ears ... we need credits !! Put some protector in and keep it screaming.

And thanks for the work getting it online !
 
As of the last update TC over took us for 3rd place. Not sure if we are gonna have enough time to catch up. But we really need all Cpu on NumberFields. AHHHHHHHhhhhh
 
Not sure what you guys think, but i moved my gpu's to asteroids this morning. 10 hours to go on City run and anandtech would need 60Million to take our spot in it which seems unlikely but they are super close to beating us on asteroids and LAF already did. hopefullly we can take back 7th in asteroids and get back 5 overall from CNT... keeping all my cpu's on NF
 
motqalden said:
Not sure what you guys think, but i moved my gpu's to asteroids this morning. 10 hours to go on City run and anandtech would need 60Million to take our spot in it which seems unlikely but they are super close to beating us on asteroids and LAF already did. hopefullly we can take back 7th in asteroids and get back 5 overall from CNT... keeping all my cpu's on NF
Click to expand...

Agree. 24 hours is a long time in this Pentathlon. What seemed like a good strategy yesterday, to defend NF, appears to be no longer viable. My eventual bunker is only going to account for a million or so credits which, even with the recent thermonuclear rigs brought to the party by fastgeek, is unlikely to bring us back to bronze.

Unless one of you chaps is sitting on a gargantuan NF bunker, IMHO I think we should be throwing every cpu/gpu resource we have at Asteroids where there is a higher potential for us maintaining or improving our position and thereby retaining our 5th overall Pentathlon place.

I could be wrong and don't want to move over unilaterally without discussion sooooo thoughts?​
 
Prime is pretty much done now and there is no way our position is going up or down in that. While its possible for us to lose a spot in the marathon. SG would need 20million points in 1 day. Not impossible but extremely unlikely. We are now 2.6+ million points behind TC in NF and i doubt they are going to slack off now as they might still beat SG at the rate they are going so we not only need a 3+million bunker, but we need another 200-300k more per hour than we are pulling currently. The single largest chance of losing ground now is in asteroids and that would be TaaT taking us over at #7. This would put us in 6th overall.
 
Gents, I'm exhausted from being at work until after 0300PST and not getting to sleep until after 0400; got back in the office at 1200. I get real cranky when I'm tired (and talk/write too much) and can't help but shake the feeling that TC "cheated" by gaming the system. Then again, maybe it all comes down to us being a "small but mighty" team being outclassed by a much larger one. None the less... I fucking hate losing and/or feeling like being fucked over... and I'm kind of angry about it.

That said. I can't spend much time on this today and am planning to leave everything where it is (pretty much all NF) unless a strong case can be made otherwise. Maybe it's not possible... but I'd really rather get back 3rd in NF. I added a lot of threads last night, and Phoenics is sitting on ~1M. The rest of our team have a lot of threads, too, just need to focus them. Could be wrong, and often am, but suspect TC blew their load already and it's (hopefully) unlikely that SG is sitting on 20+M.

GPUs on Asteroids. CPUs on NF. Hope for the best. That's my opinion.

/ramble
 
Love the spirit, but I highly doubt we can over take them. :( Don't mean to be a Debbie downer, but those guys came to play.

Looks like we may end up finishing 6th this year which, compared to last year (12th) and the year before (15th) we have come a long way. Next year hopefully we'll be stronger and a contender for a medal somewhere.
 
Well I did say it could either be a bunker or a fake hoard. Of the two, I guess it's nice it was an actual bunker vs. being bullshit.

Am still wondering how these top teams are pulling off such huge gains with bunkers on a wingman project. If they figured out a way to game the system by making them validate each others work... well... damn. We'd better figure that out, too, or we're hosed going forward.
 
BTW, I still have 3,381 pending on Asteroids and my total valid since going back online with this project is 5,999. With the massive numbers the big teams are dropping I'd still expect that pending count to be WAY lower. (All my rigs have been off this project for ~24 hours, too)
 
There was some script talking in the shoutbox; didn’t followed it.

I wonder if a script can be used to match assigned tasks within a team and make sure matching WU get done first. But that’s huge data processing as there is no simple way to get all assigned from a boinc server ; need to be polled WU by WU or list of known ID; but still in big projects with 100’000 or more WU in circulation ... it’s work

Second, easier option is that within a team people share script based the assigned WU and match then. Same result but not risking banning from project server.


That, combined with big bunker, could do quite a bit ; for wingman projects

On the weekend I will look what script they talking in shoutbox about; had no time now.
 
The latter is what I was thinking of. Somehow they're pooling data on their task/WU IDs and keeping/dropping accordingly. Now this could just be tinfoil hat talk (aka conspiracy) since I have no way of knowing if this can be scripted as such in the first place, or if there's an easy way of selectively aborting WUs via the command line... but if so, one of three things need to happen.

1) We all do it, which makes it much less lucrative.
2) The projects ban suspicious activity like this. (unlikely as they want WUs done no matter what)
3) SG / Penathalon hosts get involved and state such behavior isn't allowed (also unlikely, but maybe)

Of course 4) is "tough shit, deal". :p
 
Do older WU have a higher chance of getting validated. Lets say they pull down 1.5 weeks of WUs right when the project was announced. I suspect those old WUs have a high chance of having the wingman already returned the result by the time the person drops the WU in a bunker.
 
I'd think so, yes. I mean, look at the sheer volume of points the upper teams are churning out. One would think that this would be helping out an awful lot of people sitting on tons of pending WU's. Yet, unless my stupidly simple logical brain is missing something, that doesn't appear to be happening.
 
I think they just set their computers with a faked number of CPUs in BOINC, and a "storage" value of a few days, pulled down a "full" load of WUs enough to last them the whole competition. Set their computer back to a normal number of CPUs. Crunched them turned them in all at once Since the WUs are old ones there is a high chance they will already have a result turned in by someone else as the wingman.

Edit: atp1916 I don't think any sorted ordering is needed. They just download thousands of tasks (by reporting they have a hundred processors), then suspended networking in the client, crunched them after turning off the faked number of processors, then turned networking back on.
Edit 2: We knew what projects had a high chance of getting picked, we could have started bunkering early on all the likely suspect projects.
 
But still I get “server abort” etc for bunkered WU ... even for regular one. Increase the risk of failure too if the bunker is too big and server not contacted in times (at least I read that out of some comments form shoutbox few days ago)
 
Guys, the secret is pulling work units very, very early and pulling a lot!

When I built an asteroid bunker on my GPUs way before a GPU project was announced I had a 100% successful validated rate when I released them a few hours before their deadline.

Like Gil said. Get WUs early and get a lot.
 
I have not ran an Asteroid WU in 3 days and I'm still sitting at 200 pending validation. Literally half the Asteroid work I put in isn't going to be counted for us in the pentathlon.
 
You must log in or register to reply here.
Back
Top