9th Annual BOINC Pentathlon 2018 Strategy & Preparation

just starting 2 spot instances c3.8xlarge ; each 32 cores

Mon 7 May 22:12:31 2018 | | Processor: 32 GenuineIntel Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz [Family 6 Model 62 Stepping 4]
 
Last edited:
Landing page is here. Cost is about 50 cents an hour for a 36 vcpu spot instance and $1 per hour for a 72 vcpu. You are likely to only be able to get a 36 vcpu instance when you first sign up. You can then apply for a limit increase on the big boy.

The best availability zones are those that are unlikely to terminate your instance when a real customer comes along willing to pay full price.

In the US those would be the Oregon or N Virginia availability zones. Stay clear of Ohio. It's less expensive but highly likely to terminate the instance.
 
Do we have a guide somewhere that shows how to get things up and running quickly?? Ill buy a few slots when I get home.
 
ChristianVirtual said:
I had once storage issue with that and quite a number failed (because my default drive was only 8 GB)
Click to expand...

Good point CV. I suspect that would have been caused by the monster BHD tasks. I've just turned them off in my preferences. They may run quicker but are probably more trouble than they're worth.

Do you know how to make instances 'persistent' from the console? It would be great if I could get them to restart where they left off after termination​
 
Basically its the same as Centos ... YUM based ... you can use the same head-less setup as on regular centos.

Once you have a setup done I suggest to create a snapshot of the OS (as private AMI; Amazon Image). That you can use later to redeploy instances

I have my AMI until start of boinc (just via boinc --daemon) and the attache via console to the desired project

like
boinccmd --passwd <gui-password> --project_attach https://universeathome.pl/universe/ <weak account key>

The setup of EC2 itself:
primary you need a SSH certificate (you can make yourself) and I defined a security group to allow port SSH 22 and BOINC 31416 through the firewall.

(Sorry, no pics as they process was setup years ago; would need to redo from scratch ; but really should)
 
Last edited:
Skillz said:
Do we have a guide somewhere that shows how to get things up and running quickly?? Ill buy a few slots when I get home.
Click to expand...

I'm afraid not. CV was contemplating this but I think his brilliant work on hard-dc.com is keeping him busy.

I'm up to my eyes today before going back to work tomorrow but will try to assist if/when you get stuck. I use a Debian image whereas CV uses Centos I think.

Edit: ChristianVirtual beat me to it.​
 
I'll give it a shot when I get home from work later today. You guys be on stand buy around 6pm EST haha

Fuck those French.
 
then something like this:

Screen Shot 2018-05-07 at 22.29.43.png
 
Damm those frenchie...ergh. Full of surprises.

Free-DC (Universe) shows one entity dumping 6M.

Username is a university. Good job that they manage to "recruit" this hardware or perhaps someone may get fired or expelled?


upload_2018-5-7_7-4-13.png
 
We are gonna lose another position in Universe with this next update. China is knocking on our door.
 
you are too fast Gil. I was trying to move my post here and deleted in open forum and posted here, then you replied so i came here to delete but you had already answered :D
 
So at this point as its CPU only it doesn't seem to make sense to move anything from Universe. would be nice if we could hold off against Seti.Germany there. maybe 1 day before swimming ends or something
 
That is where the strategy comes in. Do you bunker to get a leg up on the sprint or do you continue to maintain your position and defend against others counting on you letting your guard down....
 
we need to bunker something on it. we cant just start at 0 and hope for the best. lots of points we will lose.

i am gonna bunker at least one of my 4P rigs on it.
 
By the 9th or 10th I am going to pull all my rigs off Universe and move most of them to Rosetta.
 
I'm going to move one rig to Rosetta. Not much but don't want to be far behind in sprint.

Suddenly my Universe validation rate increases. Probably due to one LAF cruncher.
All (3373) · In progress (1022) · Validation pending (697) · Validation inconclusive (0) · Valid (1339)

Should we play dirty in Universe??? Assuming you have downloaded enough tasks, download as many tasks in the last two days with the intention of not running it. Heck, I can fired up some very old laptop and store max 512 tasks and not doing anything. That would be inconsiderate, right and I'll be a bad boy?
 
Universe needs quorum of 2. If the wingman does not return the result in time, you won't get any point for your completed task. This strategy only makes sense if you can create many clients with each client holding up 512 tasks. Either via VMs or multiple BOINC instances. If a few do this, it won't work well. That's my guess.

Never try this but I recall seeing tasks held up by my wingman who did nothing and let the task expire.
 
pututu said:
Universe needs quorum of 2. If the wingman does not return the result in time, you won't get any point for your completed task. This strategy only makes sense if you can create many clients with each client holding up 512 tasks. Either via VMs or multiple BOINC instances. If a few do this, it won't work well. That's my guess.

Never try this but I recall seeing tasks held up by my wingman who did nothing and let the task expire.
Click to expand...

I have thought about this, but this only makes sense if we don't also have tasks waiting to be verfied. you could just as easily be sitting on tasks that other [H] members have already finished.
 
Man, I've had a miserable f***ing night with these f***ing computers and experimenting with different Linux OS's to find one I'm happy with. BoincOS runs on Arch and is actually very slick IMHO; but I'm not sure if there's an easy way to have it run from an actual HDD/SSD vs a thumb drive. While losing my temper with testing on the system above said box running BoincOS I managed to yank the wrong f'ing thumb drive and crashed the machine. Almost put my fist through a screen.

So after all of this crap I've only two systems online running Universe. The quad E7-8867 box and a 2P-4110. The E7 box is actually what started all of this as, for whatever reason, it just will not run at 100% load in Windows. BOINC sees all 128 threads and fires up WU's for them, but it hangs out around 65% or so load. Normally I'd just run Ubuntu as it's the path of least resistance; however this box does NOT like the new release and old ones aren't allowed to access the repos for BOINC. (Remember, I like path of least resistance in the long run) Now have it running Manjaro and it's happily pegging all 128 threads at 100%... but we'll see if it's still doing the same in the morning. Will see about getting a few more systems up tomorrow. One bad thing I can say about Manjaro is the installation takes FOREVER. Otherwise, seems to be doing the trick.

Oh, and my five problem children from Friday are all running NumberFields smoothly. Still pretty pissed off about staying so late only to have them sitting around waiting for activity before crunching any BOINC cycles. Man. WTF?
 
motqalden said:
but this only makes sense if we don't also have tasks waiting to be verfied
Click to expand...

True. Checking a few of my Universe tasks, most of the task pairs are created very close to each other except the cancelled or invalidated result which will be resent. So assuming that you have downloaded all the tasks you need much earlier, chances are it won't block those you already have downloaded earlier.

Check a few of your tasks and see the pair are created close to each other or perhaps I'm wrong.

upload_2018-5-7_21-39-20.png
 
Last edited:
I have no hope to get those 2000 WU done until end of swimming ... after that it will help in FB or I will abort and give time for Rosetta; see how it foes tomorrow also with respect to our standing
 
We don't seem to have many Rosetta crunchers compare to other teams. LAF, SG, etc have many times more than we do.

upload_2018-5-8_7-1-2.png
 
You must log in or register to reply here.
Back
Top