News Rosetta's role in fighting coronavirus

[StefanR5R]

10 work units required before the bonus starts to kick in

There is no quick return bonus at Rosetta — only at F@h. (GPUGrid has a QRB too, but not as finely grained as F@h.)

What's nice'd tasks?

In Unix and Linux, processes have a "niceness" value assigned, which determines scheduling priority. The higher the niceness = the nicer the process is towards the rest of the system, the more often it is going to be preempted by other processes which need processor time and are not as nice. -> some more info

BOINC, by default, launches the science applications as "nicely" as possible, with niceness +19.

I didn't know that the boinc service was separate from the boinc client.

They aren't really separate: Boinc client (default name of the binary: boinc) can be launched manually; but practically all Linux distributions provide launching mechanisms via their systems services facility, a.k.a. init system, which is often based on systemd these days.

Separate from the client, i.e. from boinc, are the programs which are used to control the client:

• boincmgr, a graphical control interface developed together with the client; typically also installed when the client is installed (but some distributions have it in a separate package),
• boinccmd, a command line interface which is able to show a subset of client status, and to send a subset of control directives to the client; good for scripts but also for ad hoc control from the command line; developed together with the client
• BoincTasks, a 3rd party graphical control interface, available for Windows but can also run on Linux+Wine, sets itself apart from boincmgr with much better support for simultaneous control of multiple clients
• boinctui, a 3rd party control interface with text UI, for when you need something much like boincmgr but are limited to a text terminal

Is their an optimal time beyond which useful science diminishes?

No; the Rosetta developers take care that useful work is done within any of the target CPU times which the user can choose, on slow hardware or fast hardware.

They are interested in short turnaround from when a WU is sent out to when the result comes back. But they still allow for 8 days turnaround, set in the task deadline. Sometimes they get a little impatient and create a duplicate task for an as yet unfinished WU, such that somebody else may return it earlier than the original host which may still be processing it. But both hosts will receive credit when they return a good result, regardless of which one completes it earlier. (source)

BTW, one thing which I wasn't aware of myself until now: When the target CPU time is changed at the user web page, and a project update is issued to the client or the client sends a scheduler request for other reasons, the target CPU times of all already downloaded tasks are adjusted, along with the target CPU times of yet to be downloaded tasks. (source)

Do you have a full sample config file? cc_config.xml does not exist in the RHEL install (guessing it doesn't exist in the fedora one either).

If you have a need for a cc_config.xml, then I suggest you populate it only with the options which you really need to change, and maybe a few which you don't want to change but still want to document the current values this way.

I need to go look for a complete file with defaults. In the meantime, here is the link to the documentation once more.

I would first verify that everything is working with the client:
systemctl status boinc-client

Also, if you haven't explicitly given yourself permission to run the boinc executables (i.e., you run them as root), they need to be run from /var/lib/boinc. So either cd to that directory before running or use, e.g., boincmgr -e /var/lib/boinc.

Avoid running the client as root. Remember, it downloads random stuff from the internet. Therefore, best practice is to run it with the user ID of an unprivileged pseudo-user. Virtually all distributions set it up this way by default.

The systemctl and service commands however can safely be run as root. Furthermore, it is safe to run boincmgr from your main user account.

They always have a relatively steady "ready to send" count, and it is generally much lower than the "in progress" count. Therefore I believe they are constantly generating new jobs. But I am seeing more "No tasks sent" messages in the client logs now, which I think means that the rate of requests for new work is beginning to outpace the rate of work generation. [...]
PS,
serverstatus shows 0 tasks ready to send at this moment.

Actually, it works as follows:

``The tasks ready to send are indeed the tasks ready to send by our feeder daemons. We try to keep this buffered to at least 15,000 or so whereas the actual queue that researchers submit to is not part of BOINC and is not shown to users other than the "Total queued jobs" on the R@h homepage. When this "tasks ready to send" buffer starts to reduce as it has been mentioned by the attentive participant, that means our internal queue is also low. This may happen from time to time as mentioned in my last post. These stats do not get updated frequently. The server status should be updated hourly so there may be a slowdown gathering the data as you note. The homepage status should be updated every 4 hours.``​

(source)

 

[Endgame124]

They aren't really separate: Boinc client (default name of the binary: boinc) can be launched manually; but practically all Linux distributions provide launching mechanisms via their systems services facility, a.k.a. init system, which is often based on systemd these days.

Separate from the client, i.e. from boinc, are the programs which are used to control the client:

• boincmgr, a graphical control interface developed together with the client; typically also installed when the client is installed (but some distributions have it in a separate package),
• boinccmd, a command line interface which is able to show a subset of client status, and to send a subset of control directives to the client; good for scripts but also for ad hoc control from the command line; developed together with the client
• BoincTasks, a 3rd party graphical control interface, available for Windows but can also run on Linux+Wine, sets itself apart from boincmgr with much better support for simultaneous control of multiple clients
• boinctui, a 3rd party control interface with text UI, for when you need something much like boincmgr but are limited to a text terminal

For what its worth, I ran (out of my history):

61 systemctl status boinc-client
62 service boinc-client start
63 systemctl status boinc-client
64 top
65 ls -ltr
66 tail daily_xfer_history.xml
67 cat client_state.xml
68 boinc_client > ./boinc.log &
69 top
70 tail -f boinc.log

In words - I checked the status, and saw that boinc-client was not running, so I started it and verified via status. However, top showed that there was nothing new running, so I checked the transfer history and client state xml files, which showed nothing happening. I then ran boic_client from the command line (as the boinc user) and the boinc.log started populating. The service, started by systemctl, seems to be separate from the boinc_client binary that I ran.

Once I did both, I saw usage in top and was able to follow the boinc.log to see what the client was doing.

 

[StefanR5R]

@Endgame124, top shows only some of the processes. To see all, something like

ps auxf​

works. If you know for sure for what you are looking, you can narrow it down with e.g.

ps aux | grep boinc​

In my experience, which is limited to Gentoo/OpenRC, Mint18/systemd, and openSUSE15/systemd, service boinc-client start (or its OpenRC equivalent) works reliably, service boinc-client stop (or eq.) works almost, but not entirely reliable, and by extension the same goes for service boinc-client restart.

I don't know how (and how well) it's implemented at Red Hat related distros, only that they are systemd based too.

 

[amd6502]

nice and renice command usage examples for process priority in linux

how to change process priority in linux. Nice and renice command complete tutorial with its usage and examples in linux for modifying process priority.

www.slashroot.in

What's nice'd tasks?? lol
Yea something's up with your laptop, I would've thought it'd be much quicker than your desktop, and the 2500u is supposed to have a base frequency of 2GHz.
I'm assuming you're running it off the mains, so maybe the windows power plan for when it's plugged in is virtually the same as when on battery?? Or maybe a bios power plan setting?
This for your 2500u seems oddly low & specific

Measured floating point speed

1000 million ops/se

yeah, the measured speed on my 2500u is incorrect. It's probably because I'm using a testing version of BOINC, and it's buggy. I think everything is normal. I used open HW monitor and it seems the cores are clocking at 1.4ghz on the laptop when tackling the boinc task. That's also the speed on my desktop FX when I set the ignore_nice_load flag to 1. Which makes it run silent (with <40W heat at the socket).

/sys/devices/system/cpu/cpufreq/ondemand/ignore_nice_load

If I wanted it to run at 2ghz on the laptop I think I'd have to run under the windows 'Performance' power profile. But I have no desire to spin up the laptop's fan. In fact, that's why I'm avoiding gpu boinc projects on both laptop and desktop, because those would put out an uncomfortable maximum amount of heat.

I did manage to browse both linux and windows workunit info and from what I've seen they both seem to be 80 TFLOPs normally. That at least seems to be the most common.

I'm quite satisfied with the 2500u. I can have four boinc threads running, a browser running, and another browser running under a VM, and everything is quite smooth. I tried running 6 boinc threads, and doing the same and then it got noticeably choppy. But this is pretty spectacular for me, esp'ly considering it's all running at 15W versus the 125w tdp of the desktop (which is more like 75w when you limit freq to 3ghz). So comparable multithread performance at a quarter or maybe even a fifth the wattage. .

 

[borandi]

Is there *ANY* consistency to the Rosetta credit generation?



in time taken then:

5179sec: 102 pts
5121sec: 36 pts
4921sec: 2 pts
3306sec: 68 pts
2879sec: 67 pts
1987sec: 4.6 pts

That makes zero sense

 

[biodoc]

@borandi , for some reason the benchmarks on your computer are very low.

Measured floating point speed 1000 million ops/sec
Measured integer speed 1000 million ops/sec

In boinc manager choose tools then run CPU benchmarks. It will then suspend computation on Rosetta tasks and run the benchmarks. The results with then be in your event log. Once complete, click on the projects tab then select rosetta@home and click the update button. Then check the link above and see if the benchmarks have improved.

The benchmarks on my 3950X are:

Measured floating point speed 7032.64 million ops/sec
Measured integer speed 71863.88 million ops/sec

 

[borandi]

Yes, I did. It says 'Benchmark finished in 1.9 seconds, too fast, ignoring'

01/04/2020 03:43:50 |  | Running CPU benchmarks
01/04/2020 03:43:51 |  | Suspending computation - CPU benchmarks in progress
01/04/2020 03:44:21 |  | FP benchmark ran only 1.375000 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.593750 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.765625 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.531250 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.671875 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.875000 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.406250 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.984375 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.578125 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.843750 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.578125 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.593750 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.500000 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.578125 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.609375 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.625000 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.515625 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.843750 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.671875 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.671875 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.765625 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.281250 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.625000 sec; ignoring
01/04/2020 03:44:21 |  | FP benchmark ran only 1.812500 sec; ignoring

 

[StefanR5R]

Is there *ANY* consistency to the Rosetta credit generation?

There is some, but limited consistency: Rosetta forum post

Since many of you seem interested, and mystified by the credit system at R@h, I'll try to clarify a few points. Some other projects simply use CPU seconds, or FLOPS of a task to issue credit. The problem with such a system is that it doesn't reward machines that achieve more work per second due to CPU cache, and memory available. It also depends upon the BOINC Manager to track and report the FLOPS expended on a task, and so some users began to falsify the FLOPS benchmarks and thus the FLOPS of their reported results.

To avoid these problems and design a credit system that would work going forward through new generations of hardware and capabilities, a system was adopted whereby credit granted is based on what OTHER user's machines have claimed for how difficult it is to complete each decoy (or model) of a task. The average of the results reported before you is used, rather than an arbitrary benchmark. So the credit granted does not favor a specific CPU type or operating system. It truly is based on completed work. Some tasks use different algorithms to compute their results. Assessing the effectiveness of a variety of algorithms is how the science progresses. This results in variations on computing resources required to complete them. Each type of task self-adjusts the credit it awards.

This makes a credit system that is great at reflecting the value of work to the Project Team, but makes it more difficult to benchmark your own machines. To do that well, you would have to run the same task, from the same starting seed for each benchmark run, and then compare CPU and wall-clock time, as well as the number of decoys produced. The BOINC Manager doesn't have a means of easily supporting this type of usage. Running tasks for other projects or a different class of proteins only tells you about those things. One machine might be faster than another at one type of work, and not with another type. The R@h credit system reflects the value of the outcome, without bias.

This sounds to me as if credits for new batches of tasks may need some time to converge.

Yes, I did. It says 'Benchmark finished in 1.9 seconds, too fast, ignoring'

From what I remember, this is an issue of either newer or older versions of BOINC. I don't remember which one and didn't find a source just now.

I have dual-processor PCs myself, and they don't have this problem. But I've got Linux, not Windows.

 

[borandi]

I'm on 7.14.2 I think, downloaded it the other day for this machine


Some threads that say this refer to a single core machine, and to quit any programs. This is a fresh OS install with nothing else installed. There is literally nothing else running and 100% of the CPU is free.

 

[StefanR5R]

I've got 7.8.3 on the dual processor computers.

 

[Assimilator1]

nice and renice command usage examples for process priority in linux

how to change process priority in linux. Nice and renice command complete tutorial with its usage and examples in linux for modifying process priority.

www.slashroot.in

yeah, the measured speed on my 2500u is incorrect. It's probably because I'm using a testing version of BOINC, and it's buggy. I think everything is normal. I used open HW monitor and it seems the cores are clocking at 1.4ghz on the laptop when tackling the boinc task. That's also the speed on my desktop FX when I set the ignore_nice_load flag to 1. Which makes it run silent (with <40W heat at the socket).

If I wanted it to run at 2ghz on the laptop I think I'd have to run under the windows 'Performance' power profile. But I have no desire to spin up the laptop's fan. In fact, that's why I'm avoiding gpu boinc projects on both laptop and desktop, because those would put out an uncomfortable maximum amount of heat.

I did manage to browse both linux and windows workunit info and from what I've seen they both seem to be 80 TFLOPs normally. That at least seems to be the most common.

I'm quite satisfied with the 2500u. I can have four boinc threads running, a browser running, and another browser running under a VM, and everything is quite smooth. I tried running 6 boinc threads, and doing the same and then it got noticeably choppy. But this is pretty spectacular for me, especially considering it's all running at 15W versus the 125w tdp of the desktop (which is more like 75w when you limit freq to 3ghz). So comparable multithread performance at a quarter or maybe even a fifth the wattage. .

Ah ok you didn't mention you were throttling them, or I missed it , fair enough then.
15w tdp for the 2500u? That's amazingly low! What's it's wall power draw?

 

[biodoc]

I've got 7.9.3 on my 2P but it's running linux mint 19.3. Can windows 10 pro handle a 2P system with a total of 192 threads?

 

[biodoc]

@borandi ,you could try limiting the number of threads that boinc uses to 50% (single socket) under options/computing preferences and then try rerunning the benchmark.

 

[blckgrffn]

Stupid question, maybe, but I added my Ryzen 3600 to Rosetta last night. Now it is happily crunching some work.

My 3930k this morning is now dry of work.

I should be logging into the project (which took some doing last night) withe same username/credentials on all PCs, right? I didn't supersede my older workstation by logging into a different one? TIA

 

[StefanR5R]

@borandi ,you could try limiting the number of threads that boinc uses to 50% (single socket) under options/computing preferences and then try rerunning the benchmark.

Is it predictable what Windows 10 Pro is doing when boinc is set to 50% CPUs, while the hardware consists of 2 sockets (here: 2 NUMA domains if BIOS settings are at defaults) and has SMT enabled?

Another way: Disable things in the BIOS (e.g. the 2nd processor; requires essential devices to be attached to the 1st processor), then run boinc benchmarks. If a benchmark run was successful, enable cc_config::options::skip_cpu_benchmarks. (See documentation.) Then enable things in the BIOS again.

 

[Assimilator1]

Stupid question, maybe, but I added my Ryzen 3600 to Rosetta last night. Now it is happily crunching some work.

My 3930k this morning is now dry of work.

I should be logging into the project (which took some doing last night) withe same username/credentials on all PCs, right? I didn't supersede my older workstation by logging into a different one? TIA

Yes to the 1st question , don't understand your 2nd one.
Btw, what temps is your 3600 running R@H? And what cooler's it got?

 

[StefanR5R]

I should be logging into the project (which took some doing last night) withe same username/credentials on all PCs, right?

Yes, unless you have specific reasons to earn boinc credits for different users.

I didn't supersede my older workstation by logging into a different one?

No, you didn't. The server recognizes each host as a separate one, and keeps tasks and credits and so on in separate books for each host.

Exceptions to this rule:

• Many projects have a "merge hosts" option in the web config interface, allowing the owner of two or more hosts to merge their credits.
• If certain data is copied from the boinc client data directory of one host to another, they can be made to look as one and the same host to the project server.

 

[blckgrffn]

@Assimilator1

I am running my 3600 at stock "everything" except a XMP profile on the DDR4 to bring it to it's CL16 3200Mhz spec.

I am running the Noctua nh-c12p-se14 mentioned in this thread: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/noctua-14cm-fan-replacement-needs-moar-rgb.2578356/ <-- It is not small. Historically speaking, it wasn't cheap although it was free to me. I just had to buy the ~$5 AM4 mounting bracket from NewEgg.

I was really curious after you asked the question, since I hadn't actually tested this setup since I had gotten it folding and running smoothly/fast. Ambient Temps are ~65 F in my basement. Case is a H500 ARGB.



I tweaked BOINC to hit all the cores and let it bake a little bit. This was only after a minute but I checked after 10 minutes and then even later and there was essentially no change, I just don't want to cluster this up with pics. All cores at 4+ ghz, temps hovering between 64 & 66 C. Fan noise notable but stable. No perceivable change in pitch (which I hate hate hate!). TDP looking like about 90W? I guess this just proves out the point that the only difference between a 3600 and a 3600x is the bundled cooler, as many reviews have asserted. I am not an expert at reading this program though.



This is with random ARGB fan from Microcenter too. The noctua 14cm in a 12cm space fan is about ~$18 for the "not brown" variant and would probably do even better.

I have a giant Zalman (14cm fan on that nickel plated beast) on my 3930k as well. I like the simplicity of higher end air cooling. I'll do a AIO water cooling at some point just cause (that's the "why" for the ARGB build I just did) and I am pretty happy with these results.

My 2700x with the ARGB stock cooler doesn't fair nearly as well, I don't think. I haven't tried BOINC on it because it is my work PC and it is also tucked into a cabinet and I don't want to kill it

 

[Howdy]

The Rosetta well is running dry here. How is everyone else doing?

 

[Markfw]

The Rosetta well is running dry here. How is everyone else doing?

Well, I didn;t add it up, but I still have several hundred running.

 

[blckgrffn]

The Rosetta well is running dry here. How is everyone else doing?

I was dry for a bit. 2012 memories kicked in and I set my client to get a full days worth of tasks at each opportunity

 

[TennesseeTony]

To help Rosetta maintain task supply (and also help YOU maintain Rosetta tasks in queue), one may consider logging in and changing your Rosetta preferences to run for a longer period of time. Default is 8 hours, I have always run 4 hours, but am now running at 16 hours per task.

 

[Howdy]

Default is 8 hours, I have always run 4 hours, but am now running at 16 hours per task.

running default now, was running 4hrs

They still haven't fixed the points though, 8hrs getting 8 points LOL. OTOH 8 is better than 0 or a computing error!

 

[Assimilator1]

What's the highest it's worth putting that task time to?

blkgrffn
Yeah that's a beast of a heatsink
I'm running below stock with a PPT of of 65w (according to Ryzen, thought I'd set 75w...) & hitting ~3.6 GHz, running SETI on 10 threads & F@H on the GPU it's hitting ~74C, crappy stealth HS!
When I get around to measuring up brackets for my mate to make them then my old Ultima90 will be much better (not as good as yours of course).

 

[Howdy]

What's the highest it's worth putting that task time to?

That question is above my pay grade!
Like Mr. 10 I have raised my times to 14hrs, 1 machine sitting idle as of 0400 this morning. Hopefully Rosetta can get their plans in action OR WCG fast tracks their plans to get something going on this human malware. Either way I hope to be running something to help with this current situation.