Taking over SETI rig management for my dad, settings and configurations?

Message boards : Number crunching : Taking over SETI rig management for my dad, settings and configurations?
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 5 · Next

AuthorMessage
Steve C.

Send message
Joined: 28 Sep 99
Posts: 38
Credit: 76,190,331
RAC: 107,329
United States
Message 1916928 - Posted: 4 Feb 2018, 21:18:14 UTC

So my father has been contributing to SETI for almost 20 years now, since maybe a few months after it started. I'd like to think that it had an impact on me growing up (I myself work for NASA on the Earth Observing side of things). But he's getting a bit long in the tooth so I've kind of taken over the hardware side of things for him. I'm not new to distributed computing, i did a bit of Folding back in college, and currently do some Ethereum mining of my own, and gaming.

We currently have 5 systems running, mostly CPU power but I'm adding some GPU power as well. I'd like to get some advice or direction on. I have a good mix of questions so bare with me. I've read some past threads, but most info on the topics i've searched seem to be from 2013-2015 and I'm not sure how relevant they still are to today.

The Systems:
1. Sierra-Summit
2x Intel Xeon E5440 (8 cores total, no HT)
2x 8800GT GPUs
Ubuntu 17.10 x64

2. Sierra-Prize2
2x Intel Xeon E5440 (8 cores total, no HT)
GTX 660Ti GPU (i think it's failing, all GPU units are popping out incorrect calculation errors)
Windows 7 Ult x64

3. Sierra-Summit2
2x Intel Xeon E5-2680v2 (20 cores, 40 threads)
2x GTX 760 GPUs
Windows 7 Ult x64

4. Sierra-Summit3
2x Intel Xeon E5-2697v2 (24 cores, 48 threads)
2x GTX 1050Ti
Windows 7 Ult x64

5. Sierra-Spare
2x Intel Xeon E5-2690v1 (16 cores, 32 threads)
no GPUs yet
Ubuntu 17.10 x64


Firstly, what is the time period for "RAC" is that daily average? weekly? monthly?

Next, the Sierra-Summit3 system was a relatively new build. i kind of expected the 1050Tis to be miles faster than the 760s in Sierra-Summit2. but they appear to be running the same speeds (as far as time to complete work units). both systems are setup to run 2 GPU WUs per GPU (4 total at a time). looks like the 1050Tis are running only slightly faster doing them in about 1100-1200s and the 760s doing them about 1200-1400s. is this normal for these cards? i really dont know what to expect performance wise. could this be a problem maybe of CPU resources feeding the cards instead? I have the computational preferences set to use 85% of the CPU power for CPU tasks. did i not leave enough headroom? I figured 1-2 free threads was more than enough to feed data into the GPUs.

related to the last series of questions, what is the actual meaning of the CPU/GPU numbers in the BOINC applications task list?

For example, on the Sierra-Summit system with 2x 8800GT GPUs. it's running more or less default settings. I installed ubuntu, downloaded BOINC, and just let it run at 80% CPU, and told it to use the GPUs too. in the Task list where GPU tasks are, it says something like (0.0774 CPU + 1 GPU). what is the meaning of "0.0774" here? does that mean that it's using only 0.0774 threads/cores to feed the GPU task? could this be why GPU tasks on this system take 2+ hrs to run? how do i change this value? similarly on Sierra-Summit2, before i changed the GPU to 2x WUs it was saying (0.256 CPU + 1 GPU), and Sierra-Summit3 was saying (0.502 CPU + 1 GPU). why are they all different? what is optimum?

Im just trying to maximize the work done by all of these systems. I have 3x 760s and 1x 750Ti coming (eBay) to replace the cards in the older systems and hopefully give them all a big boost.

All the help is appreciated!
ID: 1916928 · Report as offensive     Reply Quote
rob smith Special Project $250 donor
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 15643
Credit: 276,076,481
RAC: 235,672
United Kingdom
Message 1916931 - Posted: 4 Feb 2018, 21:38:20 UTC

The time base for RAC is in terms of weeks not hours or days.

The actual application being used can have a very substantial impact on the RAC achieved - some apps are fa better than others even on the same hardware, never mind across different operating systems.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1916931 · Report as offensive     Reply Quote
Steve C.

Send message
Joined: 28 Sep 99
Posts: 38
Credit: 76,190,331
RAC: 107,329
United States
Message 1916946 - Posted: 4 Feb 2018, 22:01:00 UTC - in response to Message 1916931.  

thanks for the reply.

about RAC, "weeks", OK. do you know how many weeks exactly? 1? 2? 3? or is it somehow variable and not set?

about the apps. can you be more specific? i thought the only way to crunch seti was via BOINC. is this not the case? can you give some recommendations on what "app" to use?

the two systems i compared (Summit2 and Summit3) are using the same BOINC version, on the same operating system, just different GPUs so i thought it was a fair comparison.
ID: 1916946 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4223
Credit: 240,832,150
RAC: 358,371
United States
Message 1916949 - Posted: 4 Feb 2018, 22:15:21 UTC
Last modified: 4 Feb 2018, 22:16:32 UTC

So many questions... So many answers....

RAC as far as we can see is about 2 months worth of work. That assume continuous crunching (something that been lacking as of late)

1050ti are probably only more power efficient than the 760s. (the 700 series were powerhouses but used alot of electricity) So it isn't unreasonable that they should be approximately equal.

In response to using "x" amount of CPU, I find it more useful to put 100 and then just limit the number of core via the app_config.xml This is where the argument about just what does "x" actually mean. If these are pure crunchers then I'd do this.

The 0.0774 per core is more of a "guide". I say this as if an app requires more, then it's going to use more. Cuda apps I believe still use this but the newer OpenCl SoG is better for crunching the BLC work units. (someone will bring up Petri's alpha app being faster which your linux machine can probably use) Back to the SoG app. SoG requires a full core per work unit on the GPU.

You can use the Lunatics Beta installer to install the appropriate application on your computer. Just make sure to read and click the right boxes, by default it will pick cuda 3.2 but you don't want that. Once you do that we can talk about installing an app_config.xml to limit the total number of work units (cpu+gpu) so that you can maintain some free cores for system responsiveness.

edit..

http://mikesworld.eu/download.html
ID: 1916949 · Report as offensive     Reply Quote
Steve C.

Send message
Joined: 28 Sep 99
Posts: 38
Credit: 76,190,331
RAC: 107,329
United States
Message 1916953 - Posted: 4 Feb 2018, 22:48:29 UTC - in response to Message 1916949.  
Last modified: 4 Feb 2018, 23:04:54 UTC

So many questions... So many answers....

RAC as far as we can see is about 2 months worth of work. That assume continuous crunching (something that been lacking as of late)

1050ti are probably only more power efficient than the 760s. (the 700 series were powerhouses but used alot of electricity) So it isn't unreasonable that they should be approximately equal.

In response to using "x" amount of CPU, I find it more useful to put 100 and then just limit the number of core via the app_config.xml This is where the argument about just what does "x" actually mean. If these are pure crunchers then I'd do this.

The 0.0774 per core is more of a "guide". I say this as if an app requires more, then it's going to use more. Cuda apps I believe still use this but the newer OpenCl SoG is better for crunching the BLC work units. (someone will bring up Petri's alpha app being faster which your linux machine can probably use) Back to the SoG app. SoG requires a full core per work unit on the GPU.

You can use the Lunatics Beta installer to install the appropriate application on your computer. Just make sure to read and click the right boxes, by default it will pick cuda 3.2 but you don't want that. Once you do that we can talk about installing an app_config.xml to limit the total number of work units (cpu+gpu) so that you can maintain some free cores for system responsiveness.

edit..

http://mikesworld.eu/download.html


hey thanks a lot for the info.

You'll have to excuse my ignorance a bit here, im playing catchup to grasp all the different options.

ive seen this Lunatic thing mentioned in a few other threads. is there some thread with more info about it you can link me to? what exactly IS it? is it a replacement for BOINC? or some kind of automatic configuration tool that still requires BOINC but just tweaks some settings for optimization?

all of the work units from summit2 and summit3 have SoG on them. what does "SoG" mean? i think the units coming out of Summit(1) are labelled cuda60.

like i said, all we have done was download the default BOINC 7.8.3 program then added the seti project and let it run more or less default only changing the CPU usage preference. i thought this was the only way to do this? that's what the seti official site says to do.

i did create cc_config and app_config XML files with info from some random threads i was able to find when i wanted to change Summit2 and Summit3 to use 2 WU per GPU.

i still dont understand what the 0.0774 value means. if i somehow change this value to say 0.5, what will change in terms of resource allocation?

from what i gathered on the CPU side, when i change the value in the xml file from 1 to 0.5, its some kind of weird inverse relationship where it means i run 2 jobs at once, 0.33 = 3 jobs, 0.25 = 4 jobs and so on. right?

so what happens with the CPU side? the same? and what does it mean in relation to the GPU jobs? I'm trying to understand this, I dont just want to throw random numbers into a config file and not know what i'm actually doing. because i'm not seeing the connection of how this value actually works.

but yes these are all pure crunching boxes for the most part so i'll set the value in preferences to 100% and config via the config file once i understand what all the values are doing.
ID: 1916953 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4223
Credit: 240,832,150
RAC: 358,371
United States
Message 1916959 - Posted: 4 Feb 2018, 23:38:33 UTC - in response to Message 1916953.  

BOINC is the overseeing program. A manager. It helps to prioritize which resources ( parts of the computer) get used by the different projects

The different projects then download the applications to your computers. Most of the applications used on Seti were written by the lunatics team some time ago. The installer I linked to just allows you to take direct control of your machines rather than allow the server to decide which applications to use.

SoG means Signals on GPU which will mean nothing to most of us. It's the authors term to describe what the application is doing. It really isn't important as to what it mean, just know that it uses OpenCl instead of CUDA to run.

The 0.0774 is what percentage of a CPU core is used by an application to analyze the work unit. At least that's what's it's supposed to do. Even though we say we crunch the data on the GPU, in reality there is always some portion of a CPU that is required to support it. With the older CUDA apps we could use less of a CPU core, thus why the 0.0774 value. But with OpenCl SoG they require a full CPU for each work unit.

As far as the app_config.xml. when you modify the percent of GPU from 1 to 0.5, then 2 work units will work on the GPU. If you change it to 0.33 then 3 will work on the GPU. But the percentage of CPU is still the same (with OpenCl applications) If there were cuda then they MIGHT obey the 0.0774 guidelines. But Cuda applications are slower than SoG for BLC work units.
ID: 1916959 · Report as offensive     Reply Quote
Steve C.

Send message
Joined: 28 Sep 99
Posts: 38
Credit: 76,190,331
RAC: 107,329
United States
Message 1916962 - Posted: 4 Feb 2018, 23:55:59 UTC

ok this is starting to fall in line now.

about the percentage of a "core" used number. is that logical core? or physical core? i usually refer to logical cores as threads, so i'm just trying to clarify that.

is there a rule of thumb or some guideline as to how much CPU is necessary to feed GPU units? something like .07 seems too low to me. but 0.5 to 1 seems to make more sense to me. but i really dont know. i dont want to be holding back the GPU just because im not giving it enough CPU to send the GPU units.

if i recall correctly, both Summit2 and Summit3 had different CPU core allocation for GPU units with identical settings. Summit3 was giving 0.5 and Summit2 was giving 0.25. but both of these are running SoG so maybe it doesnt matter from what youre saying?

would i benefit from switching to cuda on these machines?

is there a similar Lunatics installer for Linux? the link you sent me seems to only have windows installers.
ID: 1916962 · Report as offensive     Reply Quote
Profile Brent Norman Special Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2016
Credit: 140,649,646
RAC: 330,039
Canada
Message 1916967 - Posted: 5 Feb 2018, 0:09:21 UTC - in response to Message 1916962.  

0.5 CPU use means - reserve 0.5 CPU cores/threads (whatever you are configured for)for each GPU app. SoG apps should have 1.0 for each app.

That mikesword link will also have AVX CPU apps which I think your machines are capable of. It is in the Lunatics Windows installer, or manually installed on Linux. It is much better than stock.
ID: 1916967 · Report as offensive     Reply Quote
Steve C.

Send message
Joined: 28 Sep 99
Posts: 38
Credit: 76,190,331
RAC: 107,329
United States
Message 1916970 - Posted: 5 Feb 2018, 0:23:38 UTC - in response to Message 1916968.  

so should i uninstall BOINC to run the Lunatics installer? or run it on top of the current BOINC install?
ID: 1916970 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 3517
Credit: 212,166,734
RAC: 292,910
United States
Message 1916971 - Posted: 5 Feb 2018, 0:26:47 UTC

I don't think calling Petri's/Tbar's Linux special app an alpha is necessary. It is somewhere between a beta and a stock app for Linux, but it has much higher productivity than the projects many years old stock Linux apps.

You can read about it here.
BOINC All-In-One Build to run zi3v CUDA 8.0 in Ubuntu 12.04 to 17.04

The latest CUDA90 app is here.
Linux_zi3v-CUDA90_Special App

You can read about how to setup Linux in this Number Crunching thread.
Setting up a Linux machine to crunch CUDA80 for Windows users

The SoG app can be more productive if you install an app_config.xml file that has more aggressive tuning parameters than what the stock SoG app uses.

If you are interested in moving from the stock Linux apps to more productive versions, I would suggest popping into the NC thread I linked and read through it and post questions there. The Linux users are more than happy to assist.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1916971 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 3517
Credit: 212,166,734
RAC: 292,910
United States
Message 1916973 - Posted: 5 Feb 2018, 0:29:36 UTC - in response to Message 1916970.  

so should i uninstall BOINC to run the Lunatics installer? or run it on top of the current BOINC install?

The Lunatics Installer is supposed to gracefully install over the top of the stock BOINC folders and convert tasks as necessary.

And it has worked that way for me everytime. But I had a situation where it dumped work when I installed it over the top of BOINC when I was in a remote control session on another users computer.

So to be absolutely safe, you can finish all your work and set NNT and then install the Lunatics applications.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1916973 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4223
Credit: 240,832,150
RAC: 358,371
United States
Message 1916974 - Posted: 5 Feb 2018, 0:31:13 UTC - in response to Message 1916971.  
Last modified: 5 Feb 2018, 0:31:33 UTC

I don't think calling Petri's/Tbar's Linux special app an alpha is necessary. It is somewhere between a beta and a stock app for Linux, but it has much higher productivity than the projects many years old stock Linux apps.


It's an alpha app as it hasn't been sent to Beta site for rigorous testing and given the OK by Eric for full deployment to Main. You can make your own apps and run them but to call them "official" apps, they need the blessing of scientist of the project.
ID: 1916974 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 3517
Credit: 212,166,734
RAC: 292,910
United States
Message 1916976 - Posted: 5 Feb 2018, 0:36:19 UTC

The project has allowed anyone to compile the BOINC application code since inception. There is no requirement to submit your compilations to Beta.

Porting and optimizing SETI@home
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1916976 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 3517
Credit: 212,166,734
RAC: 292,910
United States
Message 1916977 - Posted: 5 Feb 2018, 0:39:21 UTC - in response to Message 1916974.  


It's an alpha app as it hasn't been sent to Beta site for rigorous testing and given the OK by Eric for full deployment to Main. You can make your own apps and run them but to call them "official" apps, they need the blessing of scientist of the project.

The Lunatics apps are not "official" either. But you use them yourself. The Lunatics app development has ended up as the "official" Main apps for a couple of years now.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1916977 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4223
Credit: 240,832,150
RAC: 358,371
United States
Message 1916978 - Posted: 5 Feb 2018, 0:41:53 UTC - in response to Message 1916974.  

beta and a stock app


It's not a beta, it's not a stock app. Yes, you can compile your own but you can't call it "stock". Can't call it "beta" because it has get to be released there for testing. It's an "alpha" that's being tested on Main with the consent of the scientist.
ID: 1916978 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4223
Credit: 240,832,150
RAC: 358,371
United States
Message 1916979 - Posted: 5 Feb 2018, 0:44:13 UTC - in response to Message 1916977.  


The Lunatics apps are not "official" either. But you use them yourself. The Lunatics app development has ended up as the "official" Main apps for a couple of years now.


Considering that the Lunatics team wrote the "official" application and those were tested on Beta and given the OK by Eric. Then there is the fact, most of the Lunatics installer apps are older forms of the updated "official" apps.
ID: 1916979 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 3517
Credit: 212,166,734
RAC: 292,910
United States
Message 1916982 - Posted: 5 Feb 2018, 1:07:45 UTC - in response to Message 1916978.  
Last modified: 5 Feb 2018, 1:11:59 UTC

OK, in semantics, we differ. In practical result it is good enough for Main with approximately equal Invalid/Error rate as the official Main apps. I believe only because the developer has decided to not submit the app to Beta that it hasn't been tested there. I believe he feels that is a valid choice for reasons only known to him.

The only reason that the latest Lunatics Installer doesn't have the current Main apps as default is because the previous Lunatics maintainer (Richard Haselgrove) has moved on to bigger things like the BOINC Management Team.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1916982 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4223
Credit: 240,832,150
RAC: 358,371
United States
Message 1916986 - Posted: 5 Feb 2018, 1:25:58 UTC - in response to Message 1916982.  

If Petri wants, he could always submit it to Eric for the final OK.

If Eric decides it's fine, he can upload it to the server and it will be downloaded to all linux machines along with being listed in the applications page.
ID: 1916986 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 3517
Credit: 212,166,734
RAC: 292,910
United States
Message 1916995 - Posted: 5 Feb 2018, 3:41:35 UTC - in response to Message 1916986.  

If Petri wants, he could always submit it to Eric for the final OK.

If Eric decides it's fine, he can upload it to the server and it will be downloaded to all linux machines along with being listed in the applications page.

That is absolutely true.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1916995 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4223
Credit: 240,832,150
RAC: 358,371
United States
Message 1917002 - Posted: 5 Feb 2018, 4:18:28 UTC - in response to Message 1916970.  

so should i uninstall BOINC to run the Lunatics installer? or run it on top of the current BOINC install?


Just run it on top of the BOINC, it shouldn't affect anything. Any work units downloaded will continue to show whatever application they were slated for but will instead run on the new application that is installed. All new downloads (when there is some) will have the new application shown in the Bonic Manager. An app_config.xml and Commandlines can be added later.
ID: 1917002 · Report as offensive     Reply Quote
1 · 2 · 3 · 4 . . . 5 · Next

Message boards : Number crunching : Taking over SETI rig management for my dad, settings and configurations?


 
©2018 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.