8th Annual BOINC Pentathlon

[Ken g6]

Not yet. Probably 3 more hours. They come at the end of the day..

But I have a possibly serious problem. This system has a E5-2683, and 16 gig of ram, and 2 x 980TI (which each have 6 gig of their own memory) and memory usage of the system is at 84%, 13.2 of 15.9 physical memory. But I have all these idle tasks waiting on memory !!! And CPU is at only 42% !!

WTF ???????

I suppose you need to check the BOINC memory preferences. Options -> computing preferences, Disk and Memory tab, Memory section.

Spoiler: screenshot

 

[TennesseeTony]

@Markfw

Glad we got the memory issue cleared up, and Einstein set for GPU only.

Here's a sample app_config.xml again, to go in the directory: C:\ProgramData\BOINC\projects\einstein.phys.uwm.edu

With this particular config, you will need to manually adjust your CPU usage to keep it at just under 100%. Each card should show two tasks of Einstein after you save this and then "read config files"

<app_config>
<app>
<name>hsgamma_FGRPB1G</name>
<gpu_versions>
<gpu_usage>.49</gpu_usage>
<cpu_usage>0.001</cpu_usage>
</gpu_versions>
</app>
</app_config>

 

[Markfw]

well, CPU is now 100% used, but only one video card is running, the other says "waiting to run".

CPU is set for 90% of the CPU's

Edit: I aborted all the tasks for Einstein that were not GPU, and now it says both are running, but one is at 1% utilization

edit 2. Now they are are 90% and 70% utilization....I didn't do anything

 

[Ken g6]

@Markfw are you using the cc_config.xml file? http://www.overclock.net/t/827904/how-to-multi-gpus-on-boinc

 

[StefanR5R]

I suspect use_all_gpus = 1 is default,* hence no cc_config needed normally.
It's important though to check all the various options in the computing preferences, as their defaults lean towards daily background usage, not racing mode.

Edit: *) OK, posts at Ken g6's link contradict this assumption.

 

[Markfw]

OK, I did that, and its now using a little more. Points should be up tomorrow. Especially when I get the 1080ti up (in the next few hours)

 

[Ken g6]

I just figured out how to suspend networking without VB apps noticing: Options -> Other Options, and set an invalid HTTP proxy.

 

[iwajabitw]

I just figured out how to suspend networking without VB apps noticing: Options -> Other Options, and set an invalid HTTP proxy.

Give me an example I can copy and paste in the fields.

 

[Ken g6]

Give me an example I can copy and paste in the fields.

For an HTTP proxy:
Host: 127.0.0.1
Port: 4242

Nothing else. Ensure the proxy is enabled (checkbox).

 

[iwajabitw]

For an HTTP proxy:
Host: 127.0.0.1
Port: 4242

Nothing else. Ensure the proxy is enabled (checkbox).

When I did that all of my VB Atlas tasks now say "Waiting on Network Access" after suspending the network, or do I not suspend the network after adding the proxy?

Edit: BOINC just announced

Notice from BOINC
BOINC can't access Internet - check network connection or proxy configuration.
5/10/2017 7:25:25 PM

So its all good now, Thanks

 

[Ken g6]

Wow, LHC@Home looks complicated! I think somebody mentioned the VM downloads work for itself? Also, I guess this is a balanced configuration: https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=4213

 

[iwajabitw]

I currenty have Sixtrack going, 8 at a time. Atlas uses 8 cores when it fires up. the other 4 are for the GPU running Einstein. I should have about 24hrs of LHC to bunker up.

Oh and Sixtrack doesn't use VB that I have seen. Atlas does, I think Tony mentioned it before.

 

[iwajabitw]

Do the hourly updates die off. No green or red indicators for any team the last few hours.

 

[GLeeM]

What do you put in the "hosts" file to stop Yoyo from uploading?

 

[iwajabitw]

I copied this from LHC's forums about VB errors for any one that it may help.

From: Jacob Klein
Sent: Wednesday, January 4, 2017 10:48 AM
To: Rom Walton; Marius Millea
Subject: Re: vbox debugging

I'll chime in.

Oracle VirtualBox has a "Windows process hardening" security system in their code, that will intentionally disallow VMs to run, if the security check fails. The code is a bit fragile, and when new "Windows Insiders" builds of Windows 10 get releases, they are prone to cause the problem, until Oracle fixes it. And those would lead to your problems I think.

That link Rom gave, is the best avenue to keep track of which Win10 builds work against which VBox versions.
https://forums.virtualbox.org/viewforum.php?f=38

Ultimately, what you need to do (to determine if the problem was caused by Oracle VirtualBox's "Windows process hardening"), is:
Make 2 tables.

Table 1:
Windows 10 Build Version || Minimum Oracle VirtualBox Version including the R number, which passes security check

Table 2:
VBoxWrapper Version || Maximum compatible Oracle VirtualBox Version

Building Table 1 will involve going through all those threads in that forum. Which you could do!

For instance, I know for fact that, in order to run Win10 Build 14971 or 14986, without error, you must use these VirtualBox branch versions or newer:
VirtualBox 4.3.41 r112025
VirtualBox 5.0.29 r112021
VirtualBox 5.1.9 r112020

Another for instance, is..
If you're using VBoxWrapper 26167, you must use VirtualBox 4.3.x or older
If you're using VBoxWrapper 26184, you must use VirtualBox 5.0.x or older
If you're using VBoxWrapper 26197, you must use VirtualBox 5.1.x or older

Are we having fun yet?

Regards,
Jacob

 

[Markfw]

1080TI finally arrived. Running Einstein@home. Lets see what this baby can do.

WOW, I don't know how Einstein WU's go, but it did TWO of them in 12 minutes.

How is that ?

Edit: Power, and it takes 190 watts over idle. System idle (CPU was 100%), GPU idle) was 182, Now its 372 full load everything

Edit2. And the temp is 42c@100% load, 70% fan speed.

 

[Ken g6]

for comparison, my 1060 3GB does 2 Einsteins in 31 minutes.

 

[Markfw]

I just looked. My 2 980TI's do 4 in 23 minutes, so the 1080TI is twice as fast as the 980TI

Oh, and I need to double-check, but I think it takes the same power as one 980TI.

So twice the speed at the same power.

 

[Ken g6]

When I did that all of my VB Atlas tasks now say "Waiting on Network Access" after suspending the network, or do I not suspend the network after adding the proxy?

Edit: BOINC just announced

Notice from BOINC
BOINC can't access Internet - check network connection or proxy configuration.
5/10/2017 7:25:25 PM

So its all good now, Thanks

I'm not sure this is going to work on Atlas tasks. They seem to need network communication outside the computer.

 

[StefanR5R]

I currenty have Sixtrack going, 8 at a time. Atlas uses 8 cores when it fires up. the other 4 are for the GPU running Einstein. I should have about 24hrs of LHC to bunker up.

Oh and Sixtrack doesn't use VB that I have seen. Atlas does, I think Tony mentioned it before.

SixTrack: ..................... single-threaded, no virtualbox
Atlas: ........................... multithreaded, virtualbox, said to have highest PPD
all other LHC apps: .... single-threaded*, virtualbox

*) or maybe optionally multithreaded, but that's supposed to be detrimental; I posted a link when Swimming was announced

What do you put in the "hosts" file to stop Yoyo from uploading?

Searching client_state.xml for <upload_url> shows www.rechenkraft.net. I only loaded ECM tasks, but my guess is this is applicable to the other applications too.

(I did not try upload blocking by hosts file myself though. I have yoyo on separate clients with suspended networking.)

When I did that all of my VB Atlas tasks now say "Waiting on Network Access" after suspending the network, or do I not suspend the network after adding the proxy?

Edit: BOINC just announced

Notice from BOINC
BOINC can't access Internet - check network connection or proxy configuration.
5/10/2017 7:25:25 PM

So its all good now, Thanks

I'm not sure this is going to work on Atlas tasks. They seem to need network communication outside the computer.

If vboxwrapper takes proxy settings form the boinc-client and applies them to the virtual machines of the tasks, then that won't work. But if the virtual machines take proxy settings from elsewhere, then this method would do the right thing.

(I haven't researched LHC bunkering myself, as I am not particularly enthusiastic about that sub-race, and there are the other two CPU races which I can do besides/instead of LHC bunkering.)

 

[StefanR5R]

Meanwhile, I'm downloading enough Einstein work for 2 3 days, so it'll be more likely to be completed by then.

I am also keeping "Store at least [...] days of work" high, but "Store up to an additional [...] days of work" low on my Einstein clients (2.5 and 0.1 respectively), so that I always work on somewhat old work. I have faster-than-average GPUs but prefer that the wingmen of my tasks complete theirs in the time between my download and my upload...

Edit: On my triple-GPU PC, the 2.5 days setting results in a queue of ~1000 tasks, which is limited by "not requesting tasks: too many runnable tasks". This queue really only lasts 1.5 days, not 2.5.

(Towards the end of Cross Country I will keep an eye on the queues, to avoid downloading too much work that I wouldn't finish, out of fairness to my wingmen.)

 

[StefanR5R]

[Marathon, Cosmology@Home]

Marius of C@H wrote on May 7:

"The in-progress standings of the Planck contest can now be seen here! (note: those standings include only Planck jobs, the Pentathlon counts any Cosmology@Home job)."​

[...]
Right now at C@H's own Planck app 2017 Pentathlon contest:

team rank 7 = TeAm AnandTech
user rank 7 = xii5ku​

team rank 6 = TeAm AnandTech
user rank 12 = xii5ku --- already mentally picturing Tony's taillights
user rank 13 = 10esseeTony

 

[iwajabitw]

On LHC over night none of the Atlas tasks tried to run so I don't know if the proxy settings prevent the VB from running. A bunch of Sixtracs still left to crunch from a few min per task up to 5hour tasks. Have to move my daughter back from school today so Ill be in and out as I can and TV my systems on the road. Drop time for LHC is 7pm CST, but I should be back in plenty of time for that. I have a bunch of Einstein that's getting bunkered up with LHC thanks to the proxy. New cpu parts should also be arriving today so that I can get another 280x crunching later tonight.

 

[uallas5]

So the Gaming Computer is now running GPU only, all by itself, finally.

(I thought the computer in the race was also now running GPU only, but I was mistaken, damn remote access window resolution)

 

[StefanR5R]

[Swimming, LHC@Home]

BTW, SixTrack task duration can vary a lot. They can be anywhere between <10s or >10h. The reason is that different simulations are being scheduled in the SixTrack subproject. Sometimes there will be batches issued with very short tasks, other times batches with long tasks. What you get is luck of the draw. (source in German)