8th Annual BOINC Pentathlon

[Ken g6]

For now I'm not dumping them. I'm running whatever boinc I decide to run in a VM.

 

[TennesseeTony]

Well crap. Cosmology is just like vLHC in that you can't suspend the network, or the task will suspend too!

EDIT: They have two other types of tasks, presumably not VM'd, so maybe that's the only workaround?

 

[StefanR5R]

I think I will use my Linux machines without VirtualBox in City Run and Swimming (and Sprint or Cross Country if they are CPU instead of GPU) -- and outside of those in Marathon with Cosmology's legacy application which does not require VirtualBox.

On second thought, the legacy application being run on a massive scale by many teams during the entire Pentathlon will be of questionable value to the project, from what I can tell from the FAQ. It seems to me that this is going to boil down to little more than what I have seen elsewhere called "point whoring". So, I am still not sure whether or not to run the legacy application on machines which won't have VirtualBox.

The seti-germany shoutbox says:

pschoefer: it's perfectly fine to run camb_legacy if you can't or don't want to run vbox apps
Posted: Sunday Apr. 30, 2017 at 20:08.

 

[StefanR5R]

Well crap. Cosmology is just like vLHC in that you can't suspend the network, or the task will suspend too!

Would it keep running if you block the C@H server IP at the firewall or by an "C:\windows\system32\drivers\etc\hosts" entry?

 

[TennesseeTony]

More screwy-ness from Cosmology (using VM).

Sub-Project Planck Parameter Sims likewise uses multi-threaded VM tasks, and likewise gives errors.

So that only leaves camb-legacy tasks, which Cosmology flat out tells you is of little use the them. I don't like it. I don't like it at all.

 

[iwajabitw]

I spent the morning working on Cos again. Finally got it to use virtual box properly. The tasks stayed at 0.100% for about 5-7 minutes and VB status was running, so everything was finally working. Then the ETA for the task jumped to 6 days, do what??? This is on my 5820K, so I aborted and started about 10 different tasks and they all did the same thing, 4 days being the shortest. Checked the website for the ppd and 541 ppd per task isn't worth the power to run a 4-6 day task, and there are several threads about it on there message boards. So I don't know, at a 5 day avg based on what I currently see, I will only contribute 2-4 tasks for 2000ppd for the whole time frame.

Edit: after doing some more digging I see based on our TA top producer that the ppd varies a lot. From 100-600 maybe more. The times on each work unit varies greatly also from 10K sec to over 50K seconds.

 

[StefanR5R]

As a consolation, long task runtimes make it easier to fill out long bunkering times .

Regarding low credits/task and low credits/day, that's not an issue in the Pentathlon. Similar to the Formula Boinc sprints and marathon, the team rank in each of the five disciplines is weighted by special "Pentathlon points" (from 100 to 1 for ranks 1 to 30). That way, it's fair to have a project with high credits yield in one discipline and a project with low credits yield in another. There is never going to be a direct comparison of credits between projects.

 

[iwajabitw]

As a consolation, long task runtimes make it easier to fill out long bunkering times .

Regarding low credits/task and low credits/day, that's not an issue in the Pentathlon. Similar to the Formula Boinc sprints and marathon, the team rank in each of the five disciplines is weighted by special "Pentathlon points" (from 100 to 1 for ranks 1 to 30). That way, it's fair to have a project with high credits yield in one discipline and a project with low credits yield in another. There is never going to be a direct comparison of credits between projects.

Ah, that sounds better because I have very few CPU's. But like Tony questioned earlier, how do we bunker up if we can't turn off networking for Cosmology?

 

[iwajabitw]

So it looks like this one task wont finish until the day or day after the race starts.

Edit: In the time it took to upload and get this pic posted, the time has changed to 12days!

 

[StefanR5R]

how do we bunker up if we can't turn off networking for Cosmology?

In C:\windows\system32\drivers\etc\hosts, enter a line for www.cosmologyathome.org (or whatever the project server is according to client_state.xml), pointing to an invalid IP address:
http://www.overclock.net/t/1620003/formula-boinc/410#post_26059340

Or have a firewall block access to that host.

I haven't tried either method myself yet, and whether they are proper for cosmology needs to be seen.

 

[Jondi]

More screwy-ness from Cosmology (using VM).

Sub-Project Planck Parameter Sims likewise uses multi-threaded VM tasks, and likewise gives errors.

So that only leaves camb-legacy tasks, which Cosmology flat out tells you is of little use the them. I don't like it. I don't like it at all.

It's been a while since I ran cosmology but the 'run time limit exceeded' is something I remember, the server expects you to finish the work unit in a certain time, when you don't you get that error message, presumably so not to grant you too much credit. You should find that the next work units you get given will adjust accordingly.

 

[GLeeM]

I've never run the VB app only the Legacy app.

I can only get 40 WUs and then "This computer has reached a limit on tasks in progress".

This might be enough for two days? ggrrrrhh!!

Guess I run my own goals for a bit

 

[TennesseeTony]

@iwajabitw Those run times are not at all good. Something is most seriously BOINCd there. I'm running 3 tasks at a time with 3 threads each and they finish in about 10minutes.



Our friends at OCN provided this information, (as Stefan noted earlier). I can verify that it works to block the upload, but haven't undone the mod yet to check that it uploads, don't want to lose my bunker of one task. lol
http://www.overclock.net/u/533569/10esseeTony
@10esseeTony
Code:

C:\Windows\System32\drivers\etc


Find the hosts file. Open with notepad.
Code:

127.0.0.1 localhost
::1 localhost
127.0.0.1 www.cosmologyathome.org


Make sure it looks like this. To cancel the code and allow uploads, put a # before the address.
Edited by emoga - Today at 2:53 pm

 

[TennesseeTony]

@iwajabitw Try one of these app configs, aborting all work first and not allowing new tasks, until you have restarted BOINC or told it to read config files.

3 tasks with 3 threads, leaves you 3 threads for the system and or GPU projects (I'm using this).

<app_config>
<app>
<name>camb_boinc2docker</name>
<max_concurrent>3</max_concurrent>
</app>
<app_version>
<app_name>camb_boinc2docker</app_name>
<plan_class>vbox64_mt</plan_class>
<avg_ncpus>3</avg_ncpus>
</app_version>
</app_config>

Or 2 tasks with 5 threads, leaving 2 threads for the system/GPU projects.

<app_config>
<app>
<name>camb_boinc2docker</name>
<max_concurrent>2</max_concurrent>
</app>
<app_version>
<app_name>camb_boinc2docker</app_name>
<plan_class>vbox64_mt</plan_class>
<avg_ncpus>5</avg_ncpus>
</app_version>
</app_config>

If you want a different configuration, let me know.

 

[TennesseeTony]

LOL! How funny (but to be expected), I've blocked myself from going and visiting the Cosmology website by using the 'hosts' hack to prevent uploads.

 

[iwajabitw]

I not going to take the time to screenshot it, just got home and that 1st task now says 46 days to complete. LMAO!

Edit: I'm gonna abort all tasks and watch Rogue One. I don't know whats going on, is there anything to do with VB, I let boinc load what ever it does with no additions from me.

 

[GLeeM]

I can only get 40 WUs and then "This computer has reached a limit on tasks in progress".

I asked and the admin fixed it so we can get lots more tasks!

Edit: now to figure out how to get enough GPU tasks

 

[TennesseeTony]

That's awesome GLeeM!

I have had luck on only one machine, where it simply took off and ran a VM version of Cosmology properly. The other machines (that I've gotten to so far) all required me to stop BOINC at startup, exit BOINC, uninstall Oracle Virtualbox, reinstall Oracle Virtualbox, then re-enable BOINC to run at startup and restart the PC. What a pain in the arse.

Even so, it is somewhat for nought, as the tasks run about 10 minutes and I'm not getting enough VM tasks to last one day, much less 4. But, I'll be ready once the race actually starts and bunkering isn't an issue.

 

[TennesseeTony]

Lol, good job GLeeM, now you broke their server. But before that, I was getting hundreds of VM tasks on one of the machines. :thumbsup:

EDIT: Back to normal already.

 

[TennesseeTony]

@Markfw If you wouldn't care to change your WCG preferences to run Zika only, and try to get 5 days worth of tasks, you can suspend your network again and start bunkering for the City Run for the Pentathlon. If you aren't getting enough tasks, feel free to delay until tomorrow, surely they will be back on the job and producing more tasks by then.

And Ken, I see why now you are holding your tasks, my apologies, I thought the City Run started a few days after the Marathon. Well done.

Crap, this means I have to fire up the other 8 boxes again. I'm going to have to start a go-fund-me.

 

[StefanR5R]

[Marathon, Cosmology@Home]

Memory requirements:
2 GB per job
https://www.cosmologyathome.org/forum_thread.php?id=7469

CPU limits:
A seti.germany user who experimented with several VM projects lately recommends to start at most as many VMs as there are physical cores, and optionally use remaining hyperthreaded logical cores for other projects. Restricting CPU usage of VMs in this way helped him avoid problems like stuck VMs and unresponsive machine.
https://www.seti-germany.de/forum/threads/8749-BOINC-Pentathlon-2017?p=297885&viewfull=1#post297885 (in German)

Scientific goals:
Of the three applications offered, my understanding is that the the current preference from the project's point of view is:
planck_param_sims > camb_boinc2docker > camb_legacy

They currently have a paper in peer review stage. For their response to review comments, they would like to analyze additional Planck jobs. The three top users and the top team, according to Planck jobs crunched during the Pentathlon, will get a mention in the final paper.
https://www.cosmologyathome.org/forum_thread.php?id=7468
https://www.cosmologyathome.org/forum_thread.php?id=7470

Edit:
Current (May 1, 10:00 UTC) C@H server status shows that all buffered Planck WUs have already been bunkered away. But I'm sure the admins will sort that out on short notice, seeing how responsive they are on their forums.

Edit 2:
added CPU limits

 

[StefanR5R]

LOL! How funny (but to be expected), I've blocked myself from going and visiting the Cosmology website by using the 'hosts' hack to prevent uploads.

I wish I was versed with the built-in Windows firewall. I suspect there is a dead-simple way to add one or two rules there to block only boinc from the project server.

After installation of the boinc client and starting it for the first time, I acknowledged a dialogue to allow network access by boinc. This created two inbound rules: allow all for UDP and for TCP respectively. I suspect one only needs to make a copy of that TCP rule, edit it to deny the IP address of the Cosmology server, and make sure that this rule takes precedence over the existing allow-all rule.

Edit:
Or less complicated: Simply deactivate the existing boinc rules for the time of bunkering?

 

[StefanR5R]

[City Run, WCG OpenZika]

There was a glitch in the system which alerts WCG admins when work is running low. Work buffers are being filled up again.

Edit:
"Message from server: We are currently experiencing high load and are temporarily deferring your scheduler request. Your client will automatically try again later."

 

[StefanR5R]

HOWTO: Multiple BOINC instances on the same PC

In order to build large bunkers, or in order to operate several projects simultaneously with different computing and networking preferences, it may be useful to launch two or more boinc-client instances at once on the same host.

One way to do that is to install virtual machines with guest OS (Windows or Linux) and a boinc-client instance in each.

Another, lighter-weight and IMO much easier way is to configure boinc-client (and boincmgr or boinctasks) for two or more instances on the very same OS. Luckily, mmonnin of overclock.net has written a comprehensive guide for this:

http://www.overclock.net/t/1628924/guide-setting-up-multiple-boinc-instances

Thanks mmonnin and the contributors of the OCN thread.

 

[Orange Kid]

Thanks for the great info StefanR5R.
Should be easy to do on machines with multiple drives. Just gotta wonder what it does to compute times