Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Seventeen or Bust :
Problem setting up Seventeen or Bust
Author |
Message |
|
I'm coming over from the old Seventeen or Bust project. Yesterday I got a computer started on PrimeGrid/SoB. Today I tried to get a second computer on SoB. The instructions say to download BOINC and when prompted, enter the primegrid URL, and select the subproject. This worked on the first computer but on the second one, it didn't ask for the URL and didn't give me a choice of subprojects. It put the second computer on the Reisel problem, whereas I want to be on SoB. How can I change that? | |
|
Honza Volunteer moderator Volunteer tester Project scientist Send message
Joined: 15 Aug 05 Posts: 1957 ID: 352 Credit: 6,137,993,476 RAC: 2,260,339
                                      
|
Since I can see that second computer got SoB tasks, it looks like you already figured out what was misssing...(?)
____________
My stats | |
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 14011 ID: 53948 Credit: 433,165,326 RAC: 1,015,136
                               
|
Welcome to PrimeGrid!
And also welcome to the wonderful and confusing world of BOINC venues. You stumbled into a fairly common problem.
BACKGROUND EXPLANATION:
(The nomenclature isn't terribly consistent. What I refer to as "venues" is often called "locations". If you see one or the other anywhere on the website, they're referring to the same thing. Likewise, the "<blank>" venue and the "---" venue refer to the same venue.)
BOINC doesn't let you control computers individually. Instead, you assign each computer to a venue, and then you select the sub-projects for that venue. BOINC has four venues: "<blank>", "home", "school", and "work". A while ago I modified our version of BOINC to add 10 more venues, so we have 14 total.
A computer will therefor get tasks according to the preferences you set for that venue.
New computers are assigned to the venue defined as your "default" venue. The venue to which existing computers belong can be changed at any time.
Except for the "<blank>" venue, none of the other venues exist until you explicitly create them.
If you have a computer belonging to a venue that doesn't exist, the server chooses which tasks to send.
CURRENT SITUATION:
* Your default venue is "home".
* The "home" venue isn't defined.
* The "<blank>" venue is set up to send both SoB and TRP-Sieve tasks.
* One of your computers (521171) is in the "home" venue, so the server is sending whatever it wants to send.
* The other computer (521149) is in the "<blank>" venue, so it's getting SoB and TRP-Sieve tasks, as per the settings.
STEPS TO FIX:
* Change your default venue to "<blank>". Go to the preferences page http://www.primegrid.com/prefs.php?subset=project, scroll down, and click on "Edit PrimeGrid preferences". About 7 lines from the top you'll see "Default computer location". Change that from "home" to "---". (Don't hit "update preferences" yet.)
* De-select TRP-Sieve: Scroll down a bit more than halfway on that page until you see the sub-project "The Riesel Problem Sieve (TRP-Sieve)". To the right is a box which is checked. Un-check that box. Now scroll all the way to the bottom of the page and click on "Update Preferences".
* Change the computer currently in the "home" venue to be in the "<blank>" venue. Go to your hosts page http://www.primegrid.com/hosts_user.php. You'll see both of your computers there. Notice that 521171 lists "home" as its venue while 521149 has nothing in the venue column. Click on the "details" link for 521171. At the bottom of the host details webpage is a selection box for this host's venue. Change it from "Home" to "---". Now hit "Update".
Everything should now be set up such that only SoB tasks will be sent to these computers, as well as any additional computers you add later on.
Finally, it's okay to abort the TRP-Sieve tasks. Aborting them won't hurt anything.
____________
My lucky number is 75898524288+1 | |
|
|
jMac, I welcome you to the ranks of SOB-doers. We are under-stuffed, thank you on insisting to join us. Also first post of this thread might be useful (as you are new to the PG and probably don't know how things tick here).
Thank you.
____________
I'm counting for science,
Points just make me sick. | |
|
|
Since I can see that second computer got SoB tasks, it looks like you already figured out what was misssing...(?)
Well, something straightened out. I think it lists the second computer I added as #1. The first computer I added is an I7 and BOINC Manager shows that it is running w SoB tasks. On the other computer, when I posted the message, the manager said that it was running Reisels. But now it shows SoB - and I haven't made any change. | |
|
|
jMac, I welcome you to the ranks of SOB-doers. We are under-stuffed, thank you on insisting to join us. Also first post of this thread might be useful (as you are new to the PG and probably don't know how things tick here).
Thank you.
Thanks. I started on the GIMPS project shortly after it started - when a test took about 10 hours on some sort of Pentium and you emailed in your results! About 10 years ago I switched to SoB, for three reasons (1) the Mersenne tests started taking too long (I like to see progress), and (2) SoB can be finite, and (3) to try to settle the conjecture.
When the main SoB crashed, I was #13 on the total production list. I assume that it isn't coming back, so I came here. I have nine computers to put on it (three i7 and six i5).
One good thing about this software is that it must not use the GPU and isn't overheating my CPU. On one of my i7s, I could run only two threads of the GIMPS software or it would make my CPU too hot. Other computers I had to limit to two or three threads. I have eight threads running on it now, and the CPU temperature is fine! | |
|
|
STEPS TO FIX:
Thanks, I did that, and I will be adding more computers. One question: one i5 only has 4GB whereas the others all have at least 12GB. Is 4GB enough? | |
|
RafaelVolunteer tester
 Send message
Joined: 22 Oct 14 Posts: 911 ID: 370496 Credit: 550,234,793 RAC: 442,833
                         
|
One good thing about this software is that it must not use the GPU and isn't overheating my CPU. On one of my i7s, I could run only two threads of the GIMPS software or it would make my CPU too hot. Other computers I had to limit to two or three threads. I have eight threads running on it now, and the CPU temperature is fine!
That's probably because you have HT on, which causes such a huge RAM bottleneck that the CPU cores can't get enough work done to warm up.
We always advise turning HT off if you intend to do LLR testing. Or at the very least, only run half of the available threads. Do so and your tasks should go a bit over twice as fast; in other words, you'll be doing more crunching and getting results faster.
Thanks, I did that, and I will be adding more computers. One question: one i5 only has 4GB whereas the others all have at least 12GB. Is 4GB enough?
The problem isn't the size of the RAM. Rather, the bandwidth: LLR needs LOTS of it. So while 4GB would be plenty, the lack of dual channel (and going even further, the fact that it's most likely a single rank stick) would hurt it pretty bad compared to something like 2x4 or 1x8 dual rank. OCing RAM frequency, if your mobo supports it, is also something that'll boost perofrmance. | |
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 14011 ID: 53948 Credit: 433,165,326 RAC: 1,015,136
                               
|
STEPS TO FIX:
Thanks, I did that, and I will be adding more computers. One question: one i5 only has 4GB whereas the others all have at least 12GB. Is 4GB enough?
Yes, that's plenty. The main memory requirements for most tasks are minimal, and are similar to what you would need with the Seventeen or Bust client. At the present time, all BOINC tasks at PrimeGrid use less than 250 MB. Some use a lot less; The SoB tasks use about 125 MB.
____________
My lucky number is 75898524288+1 | |
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 14011 ID: 53948 Credit: 433,165,326 RAC: 1,015,136
                               
|
I think it lists the second computer I added as #1.
Computers are listed on that page sorted by the most recent time each computer contacted the server. Currently active computers will therefor be constantly switching positions in the list as each of your computers needs to talk to the server to get new work or report results.
____________
My lucky number is 75898524288+1 | |
|
|
I got 40 CPUs running SoB the other day, but now I want to switch to the Riesel Project. I've added the Riesel project, but if I deselect the SoB project, will the tests in progress complete? (I don't want to waste what I've done.) | |
|
RafaelVolunteer tester
 Send message
Joined: 22 Oct 14 Posts: 911 ID: 370496 Credit: 550,234,793 RAC: 442,833
                         
|
I got 40 CPUs running SoB the other day, but now I want to switch to the Riesel Project. I've added the Riesel project, but if I deselect the SoB project, will the tests in progress complete? (I don't want to waste what I've done.)
The preferences you set only affect yet to be downloaded WUs. Anything you currently own won't stop. You can change it at will without fear of losing your current work.
Also (just as a heads up in case you don't know these):
1- We strongly recommend disabling HyperThreading on any machine running LLR tests. Alternatively, only running half of the available threads, aka the number of actual cores. But disabling HT is prefered. Due to RAM and cache limitations, especially on a project as big as SoB, HT actually decreases performance rather than boosting it.
2- Speaking of which, due to (again, cache and RAM limitations), it's adviseable to not run all of your cores on a single big project like SoB. Your mileage may vary, but my personal rule of thumb on a quad core is 2 SoBs + 2 small projects (sieves, PPSE, SGS, other Boinc Projects) or 1 Sob + up to medium sized projects (MEGA, PPS, TRP, SR5).
3- TRP SV and TRP (LLR) are 2 different things. The first is a sieve to remove candidates; it will never find primes, nor prove the conjectures; on the flipside, it's not as intensive on the CPU, and the tasks are shorter as well. The second is pretty much the opposite.
4- Looking at your WUs, seems like you have your Boinc caching preferences set to store lots of WUs at once, no? We recommend running both "Store at least" and "Store up to an aditional" at 0 or 0.something when crunching for Primegrid. | |
|
|
Also (just as a heads up in case you don't know these):
1- We strongly recommend disabling HyperThreading on any machine running LLR tests. Alternatively, only running half of the available threads, aka the number of actual cores. But disabling HT is prefered. Due to RAM and cache limitations, especially on a project as big as SoB, HT actually decreases performance rather than boosting it.
That seems to be true on one of my i7s but not the other. My i5s running four SoB tests are taking about 10 days. One one i7 running eight SoB tests, they are taking 20-23 days (less throughput) but on the other i7 they are taking 14-16 days (more throughput than running four at a time on an i5 but I don't know how it compares to running four on an i7.
Also, I don't see a place to disable hyperthreading.
4- Looking at your WUs, seems like you have your Boinc caching preferences set to store lots of WUs at once, no? We recommend running both "Store at least" and "Store up to an aditional" at 0 or 0.something when crunching for Primegrid.
Yes, on some of the computers I changed those to a few days so if my internet is out for a while, it will have something to do. | |
|
|
You can set different computers to different "locations" (specifically: "default", Home, Work, School, Sun, Mercury, Venus, Mars, Jupiter, Saturn, Uranus, Neptune, and Pluto) so you can have different settings for different computers/groups of computers. I would set all your i7's to a separate "location", then go into the "Computing Preferences" in the Primegrid website and change "use number of CPU's" to 50% (only for your i7 location). This way you can also have separate "Primegrid preferences" (i.e. which subprojects to run), so if you want to run different tasks on different computers you can.
Alternatively you can physically turn off hyperthreading, I don't know how to do this as I don't actually have any i7's but I believe it's something in the BIOS. | |
|
|
Some BIOS manufactures lock Hyperthreading on. In this case under computing preferences use the following: On multiprocessors, use at most (Enforced by version 6.1+) 50% of the processors | |
|
|
You can set different computers to different "locations" (specifically: "default", Home, Work, School, Sun, Mercury, Venus, Mars, Jupiter, Saturn, Uranus, Neptune, and Pluto) so you can have different settings for different computers/groups of computers. I would set all your i7's to a separate "location", then go into the "Computing Preferences" in the Primegrid website and change "use number of CPU's" to 50% (only for your i7 location). This way you can also have separate "Primegrid preferences" (i.e. which subprojects to run), so if you want to run different tasks on different computers you can.
Alternatively you can physically turn off hyperthreading, I don't know how to do this as I don't actually have any i7's but I believe it's something in the BIOS.
I don't want to turn hyperthreading off because I need it when running some of my own programs. I haven't set up different locations, but I changed the i7s to use 50% of the CPUs. Looking at the task manager, resource manager, each of the eight threads is used about 50% so I can't really tell if it is assigning one physical CPU per task.
| |
|
RafaelVolunteer tester
 Send message
Joined: 22 Oct 14 Posts: 911 ID: 370496 Credit: 550,234,793 RAC: 442,833
                         
|
That seems to be true on one of my i7s but not the other. My i5s running four SoB tests are taking about 10 days. One one i7 running eight SoB tests, they are taking 20-23 days (less throughput) but on the other i7 they are taking 14-16 days (more throughput than running four at a time on an i5 but I don't know how it compares to running four on an i7.
Also, I don't see a place to disable hyperthreading.
Hint: I5s and i7s aren't directly comparable to each other due to the extra L3 cache. Believe it or not, it helps. RAM bandwidth differences between systems also matters. The lack of FMA3 (present only in Haswell and above) only makes comparisons even more difficult. Not to mention that due to there being 6 different Ks being crunched, Boinc WUs aren't even reliable to begin with.
Sound like a lot? Cuz it sure does to me!
Also, I don't see a place to disable hyperthreading.
It's in the Bios. Get to it and fiddle around, you should find it somewhere. | |
|
|
Hint: I5s and i7s aren't directly comparable to each other due to the extra L3 cache. Believe it or not, it helps.
I know it does.
Something strange is going on - an hour or two ago I changed the i7s to use 50% of the CPU. Now there are four tasks running, but the time remaining on them is increasing! | |
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 14011 ID: 53948 Credit: 433,165,326 RAC: 1,015,136
                               
|
Hint: I5s and i7s aren't directly comparable to each other due to the extra L3 cache. Believe it or not, it helps.
I know it does.
Something strange is going on - an hour or two ago I changed the i7s to use 50% of the CPU. Now there are four tasks running, but the time remaining on them is increasing!
BOINC often gets confused about run-times.
____________
My lucky number is 75898524288+1 | |
|
RafaelVolunteer tester
 Send message
Joined: 22 Oct 14 Posts: 911 ID: 370496 Credit: 550,234,793 RAC: 442,833
                         
|
Hint: I5s and i7s aren't directly comparable to each other due to the extra L3 cache. Believe it or not, it helps.
I know it does.
Something strange is going on - an hour or two ago I changed the i7s to use 50% of the CPU. Now there are four tasks running, but the time remaining on them is increasing!
BOINC often gets confused about run-times.
*Cue that SoB result that was throwing statistics off with it's mighty 2.5million hours as cpu run-time.*
I just re-saw the thread the other day, that's why I remembered it. | |
|
|
BOINC often gets confused about run-times.
That seems to be true. I started these SoBs 3 days ago. The i5s are saying 5 days remaining. The i7s, even after cutting it down to 50% of the CPUs, are showing 10-12 days remaining (despite higher clock speeds and more cache). Prime95 would tell you the iterations per second it was getting, and I would use that to optimize the throughput.
| |
|
|
Hint: I5s and i7s aren't directly comparable to each other due to the extra L3 cache. Believe it or not, it helps.
I know it does.
Something strange is going on - an hour or two ago I changed the i7s to use 50% of the CPU. Now there are four tasks running, but the time remaining on them is increasing!
BOINC often gets confused about run-times.
*Cue that SoB result that was throwing statistics off with it's mighty 2.5million hours as cpu run-time.*
I just re-saw the thread the other day, that's why I remembered it.
https://cdn.meme.am/instances/500x/69725100.jpg | |
|
mackerel Volunteer tester
 Send message
Joined: 2 Oct 08 Posts: 2645 ID: 29980 Credit: 568,565,361 RAC: 266
                              
|
Setting to use 50% of processors doesn't always work perfectly. If using Windows, it sometimes puts two threads on the same core still. You can work around this somewhat by manually setting affinity for boinc and or SoB tasks. Setting it for boinc will make it apply to new tasks, so any already running also needs to be manually set once. Note this does not persist between restarts.
Setting affinity also provides a small performance increase relative to not doing so, and makes it near enough equal to turning HT off. | |
|
|
Setting to use 50% of processors doesn't always work perfectly. If using Windows, it sometimes puts two threads on the same core still. You can work around this somewhat by manually setting affinity for boinc and or SoB tasks. Setting it for boinc will make it apply to new tasks, so any already running also needs to be manually set once. Note this does not persist between restarts.
I use one of the i7s for my main computer and I have it set to suspend until after 1 minute of inactivity. If I set the affinity of BOINC to the even-numbered CPUs, when they start running again, with that assign one task to each CPU?
-----
I let the computers run uninterrupted for about 95 minutes. The time remaining doesn't seem to be accurate, so I looked at the percentage of progress.
The 3.1GHz i5-2400 was doing 0.0063%/minute on SoB.
The 4.0GHz i7-4790K (set to use 50% of CPUs) was doing 0.0093%/minute.
So the i7 was 1.48x the i5, which is about what I expect.
| |
|
RafaelVolunteer tester
 Send message
Joined: 22 Oct 14 Posts: 911 ID: 370496 Credit: 550,234,793 RAC: 442,833
                         
|
The 3.1GHz i5-2400 was doing 0.0063%/minute on SoB.
The 4.0GHz i7-4790K (set to use 50% of CPUs) was doing 0.0093%/minute.
So the i7 was 1.48x the i5, which is about what I expect.
I'd actually expect even more due to FMA3... Not to mention higher clocks and better IPC. | |
|
|
I have it set to suspend until after 1 minute of inactivity
You can do this if you want, but all BOINC projects automatically run with the lowest processing priority, so there shouldn't be any noticeable slowdown when using the computer at the same time as crunching. I prefer the "suspend BOINC when CPU usage is above a certain percentage" option, just in case I'm doing something else processor-hungry, but in theory even that shouldn't be necessary.
If I set the affinity of BOINC to the even-numbered CPUs, when they start running again, with that assign one task to each CPU?
I just tested it; yes, if you have "Leave suspended tasks in memory" OFF, then the processes will quit and relaunch every time you suspend/resume, so your affinity settings should take effect as soon as you resume. If you have the option on, obviously your processes will only quit when you shut down the computer, so make sure to have the option off. Remember that, at least according to mackerel, you'll have to set the affinity every time you reboot. | |
|
mackerel Volunteer tester
 Send message
Joined: 2 Oct 08 Posts: 2645 ID: 29980 Credit: 568,565,361 RAC: 266
                              
|
In LLR tasks, the peak IPC of Haswell over Sandy Bridge is about 50% extra (or 1.5x that of SB). Ram speed is most likely limiting it far below the above. Even dual channel 2400 would be rather limiting. | |
|
|
In LLR tasks, the peak IPC of Haswell over Sandy Bridge is about 50% extra (or 1.5x that of SB). Ram speed is most likely limiting it far below the above. Even dual channel 2400 would be rather limiting.
The 1.5x corresponds to about what I get with my programs that are CPU intensive and usually memory inensive. | |
|
|
The percentages on the Manager, Tasks tab are not updated very often. However, one one of my i5s, I've had four SoBs running for 4 days, 22 hours, and about the same number of minutes. All of the exponents are about the same size. However, their progress is 63.2%, 63.1%, 53.9%, and 52.9%.
Why are there such big differences? (The 63.2% and 53.9% are even running the same value of k.) | |
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3233 ID: 50683 Credit: 151,443,349 RAC: 99,549
                         
|
The percentages on the Manager, Tasks tab are not updated very often. However, one one of my i5s, I've had four SoBs running for 4 days, 22 hours, and about the same number of minutes. All of the exponents are about the same size. However, their progress is 63.2%, 63.1%, 53.9%, and 52.9%.
Why are there such big differences? (The 63.2% and 53.9% are even running the same value of k.)
Because all cores are not used by OS at same percentage. Some core is used more, some is used less. Only on computer that doesnot have any other load you will find similar % at similar time per core. I never manage to get that under Windows, but under Linux, I can . But those machines under Linux do only one thing: LLR.
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! | |
|
|
Because all cores are not used by OS at same percentage. So some is used more, some is used less. Only on computer that does not have any other load you will find similar % at similar time per core.
There is very little other load on that computer. If I bring up Windows Task Manager and wait a few seconds, all of the cores are showing 25.0%, except that once in a while one will drop to 24.8 or 24.9.
If I cut it down to using at most 30% of the cores (i.e. one core), then it spreads the task over all cores, so it isn't assigning one core to one task. Perhaps that is the reason. | |
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3233 ID: 50683 Credit: 151,443,349 RAC: 99,549
                         
|
Does you surf on that PC?
Open and close browser?
Load pages in browser?
All those action "eat" CPU cycles...
Also SoB tasks are huge ones, so they dont fin in L3 cache.. and CPU is oversaturated. THat little difference is multiply with time, and that is also reason of different percentage done per core.
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! | |
|
|
Does you surf on that PC?
Open and close browser?
Load pages in browser?
None of that.
| |
|
axnVolunteer developer Send message
Joined: 29 Dec 07 Posts: 285 ID: 16874 Credit: 28,027,106 RAC: 0
            
|
Might be different FFT sizes. | |
|
|
Memory contention may also be an item. With SOB size, you are moving numbers in and out of memory a lot. Windows is not very efficient in which core it gives priority to for that contention, so you could get timings that differ. There is also other background items that always run, or in a case of virus scanner, sometimes run. These will take up cycles. Even BOINC Manager takes a few minor cycles. All these cycles add up over time. | |
|
|
A great recommendation for controlling affinity, priority and many other settings is Process Lasso. You can set the LLR app to use only physical cores without turning HT off and will remember your settings even after a reboot. Plus much, much more.
Here's link to DL page:
https://bitsum.com/?inproduct
I'd recommend this to most crunchers here at Primegrid because it does so much in optimizing most all systems, not just ones with HT.
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713
| |
|
|
A great recommendation for controlling affinity, priority and many other settings is Process Lasso. You can set the LLR app to use only physical cores without turning HT off and will remember your settings even after a reboot. Plus much, much more.
I downloaded it and it seems quite sophisticated. It may be beyond my understanding, so some help with the settings may be needed.
I'm looking at ProBalance - default process CPU affinities. It asks for the name of a process and you can assign it to any CPU. However, all of the instances of PrimeGrid have the same name, PrimeGrid_cllr.exe, so I don't see how to assign each one to its own CPU.
P.S. - I see how to do it now. In the list of processes, right-click on the instance's affinity column. | |
|
Message boards :
Seventeen or Bust :
Problem setting up Seventeen or Bust |