Question Linux BOINC help

[Fardringle]

I've been AWOL for a while, mostly because I lost most of my "herd" and have only had my one home PC running BOINC since early last year.

I recently got a new work laptop and purchased my old retired Dell E6430 laptop for $5. It's not a powerhouse by any means, but should still be useful for some things so I thought I'd put it to work to help maintain my favorite silly stat of getting the most milestones in the most projects. I love seeing all of those 1's in my Free-DC milestones list!



Specifically, there is one project called Wanless2 (WEP-M+2) that has been mocking me with a zero in my stats for years because it doesn't have a Windows client and all of my computers have always been Windows only. I installed Linux on this laptop and fired up BOINC, which was pretty painless, but the benchmarks and run times seem to be ridiculously wimpy compared to other similar systems running this project. I've never used Linux so I don't even know where to start to try to troubleshoot it to see if there are missing drivers, bad distro, or whatever. The laptop is Computer #35327 and it has an i7-3720QM CPU @ 2.60GHz with 8 GB of RAM. Its benchmarks of 1750.32 float and 10353.41 integer are a LOT lower than any other computer that has validated the same tasks, including some that should not be significantly faster, like an i5-2400 (not K) with 3304.33/19931.46 benchmarks (and 1/2 to 1/3 faster WU processing times). I checked some benchmark sites for comparison. cpu.userbenchmark.com says that my i7 is equal or slightly faster than the i5 in benchmarks, and Passmark CPUBenchmark.net says my i7 should be a LOT faster, but BOINC apparently does not agree...

I plan to run the benchmarks for those two sites on the laptop tonight (if they will run in Linux) to see if the results are really low or not.

In the meantime, is it possible that the ChaletOS distro (based on Ubuntu) that I used simply isn't efficient or isn't using the hardware properly? If so, what distro would you recommend?

I guess it's also possible that I'm overheating the laptop by running 8 BOINC tasks on it, but it doesn't seem to be particularly warm, and I don't know what programs are available to monitor the actual temperatures in Linux.

 

[StefanR5R]

It has been brought up in this forum and elsewhere that the mathematical background of Wanless2 is not sound. Having no points there may be considered a virtue.

 

[Fardringle]

It has been brought up in this forum and elsewhere that the mathematical background of Wanless2 is not sound. Having no points there may be considered a virtue.

I'm honestly not concerned about the validity of the work in that project. I have a handful of projects that I really support. The rest just get attention long enough to make funny numbers on my Free-DC stats page..

I'm mostly just interested in finding out why my laptop has significantly lower BOINC benchmark scores (and related WU times in Wanless2) than other machines with similar CPUs. I'm hoping I can fix it, both to get the milestone in want in Wanless2 quicker and to make it more effective when I move it to other projects.

 

[TennesseeTony]

Try Linux Mint, several here use that, and you will have a better support structure.

Also, there is a way to force BOINC to redo those benchmarks, and then turn off the benchmarking at every restart, but how escapes me at the moment.

 

[Fardringle]

I'll try Mint to see what happens.

And I did manually re-run the benchmarks several times. The numbers were pretty much the same every time. Hopefully it's just the ChaletOS distro being inefficient.

 

[StefanR5R]

@Fardringle,
do you get different benchmark numbers if you set BOINC to for example "Use at most [ 12.5 ] % of the CPUs"?

How do your processor clocks look when you run the benchmark/ when you run some actual tasks?
You can check for example with

cat /sys/devices/system/cpu/cpufreq/policy*/scaling_cur_freq​

Apropos, maybe this could be useful info too:

grep -r . /sys/devices/system/cpu/cpufreq/policy0/​

You may be able to get sensors data with the simple command

sensors​

On a newly set up Linux system, it may be necessary to run

sensors-detect​

as root user once in order to configure the drivers which the sensors command will be accessing, depending on distribution.

@TennesseeTony,
<cc_config>

<options>​

<skip_cpu_benchmarks>1</skip_cpu_benchmarks>​

</options>​

</cc_config>
inhibits automatic and manual benchmarking runs.

 

[Fardringle]

I'll attempt to answer those questions when I get home tonight.

 

[Fardringle]

Limiting BOINC to a single core (12.5% of CPU time) gets a benchmark of 4530/22710. Running the benchmark with 4 cores active (50% of CPU) gets 3888/18857. So maybe HT on this CPU (or this Linux distro) isn't very efficient...?

I'm not sure how accurate the SENSORS command is. With BOINC running on just one core, it says the CPU temperature is 101 C. Even if that wasn't almost melting temperature I question the validity since if I snooze BOINC and then immediately run SENSORS again, it says the CPU temperature has dropped to about 50 C in 2-3 seconds. That's pretty unlikely, but it may be that the CPU is throttling because the system thinks that it is overheating.

Running the command to check CPU frequency while the system is totally idle gives 8 lines with numbers ranging from 1200062 to 1279382. Running it again with BOINC fully active gives 8 lines with numbers ranging from 3199929 to 33999906. I don't know how to "translate" those numbers.

Edit: Out of curiosity, I disabled HT in the BIOS and ran the tests again. SENSORS says the CPU temperature under full load is hovering around 85-87 C, and the BOINC benchmark (using 100% CPU) is 3527/19323, so it looks like either the system is overheating running 8 tasks, or years of abuse have somehow broken the HT function in the system. Also, Wanless tasks are estimating about 90 minutes to complete instead of the 4-5 hours that they have been taking.

Maybe I'll just run it with 4 cores.

 

[Fardringle]

Yep, I'm going with four cores. Running 8 cores and getting the bad benchmarks before I started this thread, the computer was averaging 3.8-4.8 hours per WU and getting 25-28 points for each WU. Running 4 cores overnight with the much better benchmarks, it is averaging 48-50 minutes per WU for the same 25-28 points each.

That's a jump from about 1070 to about 3120 per day. I'll take it!

Thanks for your help! I still might switch to Mint Linux later on, but for now I'll leave the system as is and see how it performs over a longer time period.

 

[StefanR5R]

I'm not sure how accurate the SENSORS command is.

Accuracy is a question for the sensor hardware.
As for the software, it is rather a question of correctness. If these sensor data are read from the Ivy Bridge CPU itself (not from an external sensor which in turn is attached to some sketchy 3rd party instrumentation chip), then the kernel driver is probably getting it right.

With BOINC running on just one core, it says the CPU temperature is 101 C. Even if that wasn't almost melting temperature I question the validity since if I snooze BOINC and then immediately run SENSORS again, it says the CPU temperature has dropped to about 50 C in 2-3 seconds.

If it is from the CPU's own sensors, then there is not much heat capacity involved. Still, it is certainly unexpected that an 8-thread workload would be so much worse than a 4-thread workload.

That's pretty unlikely, but it may be that the CPU is throttling because the system thinks that it is overheating.

Given the bad task runtimes, my money is on some sort of thermal throttling too.

Running the command to check CPU frequency while the system is totally idle gives 8 lines with numbers ranging from 1200062 to 1279382. Running it again with BOINC fully active gives 8 lines with numbers ranging from 3199929 to 33999906. I don't know how to "translate" those numbers.

These should be the core clocks in kHz:
1,200,062...1,279,382 kHz idle = 1.20...1.28 GHz idle
3,199,929...3,399,??? kHz loaded = 3.20...3.40 GHz loaded

 

[TennesseeTony]

I assume you would have already checked to see if the fan runs, and/or used a can of air/vacuum hose on the vents? Still a heck of a deal at $5.

 

[Fardringle]

I assume you would have already checked to see if the fan runs, and/or used a can of air/vacuum hose on the vents? Still a heck of a deal at $5.

Yes, the fan runs, and I have cleaned it out, but I haven't tried replacing the thermal paste yet, mostly because I don't have any at the moment. I need to get some more..

Yes, it was definitely worth the $5 to bring it home. Most of the time we recycle or donate the old machines when they are replaced, but I decided to keep mine for the "processing fee" just to see if I could do anything useful with it at home. It's technically free, but the "fee" makes it count as a purchase instead of a gift or an employee bonus that would create a tax/payroll hassle.

 

[Fardringle]

Thank you for the info, Stefan. I didn't feel an excessive amount of heat when running 8 threads, but it does feel cooler only running 4, and the fan is not running at 100% all the time. Hopefully new thermal paste will resolve that issue, but maybe the limited cooling in the laptop just can't keep up with 8 threads running at 100% all the time.

It's definitely doing better in BOINC with HT turned off...

 

[Fardringle]

I finally had a bit of time and decided to open up the laptop to take a peek inside. There was NO thermal paste on the CPU or chipset! Not even crusty dry old stuff...nothing at all there...

I wiped the chips and the heat sinks with alcohol and applied a bit of Artic Silver on them. I'm running 8 threads in BOINC now and the CPU temps are bouncing around between 78 C and 88 C, which is still a bit high but nowhere close to the 101 C that the sensors command was reporting previously.

The BOINC benchmark score is lower with HT enabled and 8 threads running than it was with just 4, but still a lot better than the scores it was reporting when I first started playing with this system, and BOINC is estimating around the same 50 minutes completion times for Wanless2 tasks that it has been getting when running 4 tasks, so hopefully that's a good sign. If it can maintain around the same average WU times while running twice as many WUs, that will make me happy!

I'll let it run like this for a while and see how it does.

 

[Fardringle]

After about 24 hours, it looks like the laptop is averaging around 72-74 minutes for tasks worth 28-30 points. That's a bit longer than the tasks were taking running on 4 cores, but less than double the time and worth a few more points, so potentially about 4500 points per day. I'll have to wait and see what it actually gets since the potential PPD before was about 3100 per day and it was actually getting just under 3000. If it gets anywhere close to the estimated 4500, that's a pretty good additional improvement and will get this project to my milestone goal quicker so I can move on to a project that isn't lame..

 

[Fardringle]

Averaging 4291 points per day after I added the Arctic Silver. It's a tiny drop compared to the numbers that high end GPUs can do in other projects, but it's definitely a solid improvement in this silly and nearly worthless project!

I guess the moral of this story is that a good cleaning and new thermal paste should ALWAYS be the first step when putting an old computer to use.

 

[Fardringle]

3 week update, just for an excuse to post something..

The Sensors command says that the CPU temperature hasn't ever been over 87 F since I replaced the thermal paste, so that's acceptable. Daily average is 3400 points per day in the Wanless2, including a couple of bad days when my daughter kicked the power adapter out of the laptop (it's under my desk) and I didn't notice right away. It's about 4K per day without those no production days.

Not great compared to anything remotely modern, but it's steadily getting closer to my stats goal for the project so I can have the laptop start doing something else. I'm thinking I might let it run RC5, OGR, or GIMPS since I've never done those as it would mean having to turn off BOINC on my main PC for a while and I don't want to do that. Folding@home would fit that category as well, but from what I've read it needs a good GPU to be effective, and that's something this laptop definitely does not have...

 

[Ken g6]

I wouldn't run GIMPS on a laptop. It would get too hot. Except maybe if you do trial division.

RC5 has a BOINC wrapper in Moo Wrapper. It also is best on a GPU. You can link your d.net and Moo Wrapper accounts.

OGR has a wrapper in Yoyo@home. It doesn't link to d.net, so you might want to do that separately.

 

[Fardringle]

Good info on GIMPS. Maybe I'll have to devote some time to it on my PC eventually. I mostly just want to do a bit on each of those projects to add to my "projects with results" stats on Free-DC, so while I do have results in Moo Wrapper and Yoyo@home, I don't have any yet for the non-BOINC RC5 and OGR and folding@home projects.

 

[StefanR5R]

RC5 has a BOINC wrapper in Moo Wrapper. It also is best on a GPU. You can link your d.net and Moo Wrapper accounts.

To add to this:

• If you created a distributed.net account before a moowrapper account, you can link the latter to the former by changing the distributed.net ID at the moowrapper project preferences.
• If your distributed.net account is not yet linked with d.net's team AnandTech 10635, you can do so in a few simple but not entirely intuitive steps.
• The distributed.net ID is normally an e-mail address which is visible to everyone. If you rather have a pseudonym or your real name instead of the e-mail address exposed there, you can change that at distributed.net as well.