how can i run multiple BOINC clients on the same host?

[Sunny129]

like the title says, i would like to know how to run multiple instances of the BOINC client on the same host. i've been following the instructions laid out in Vyper's Blog, which i'm sure some of you might already be familiar with. basically, it tells you to do the following:

1) create a secondary BOINC data directory using a name and a location of your liking (i chose g:\ProgramData\BOINC_2 as the name/location of my other BOINC data directory).
2) using the command prompt, enter “c:\Program Files\BOINC\boinc.exe” –allow_multiple_clients –dir g:\ProgramData\BOINC_2, which causes BOINC to place all the necessary files in this directory (hit ctrl+c to break it).
3) open the gui_rpc_auth.cfg in the new BOINC data directory and change the password to something easy to remember.
4) using the command prompt, enter “c:\Program Files\BOINC\boinc.exe” –allow_multiple_clients –dir g:\ProgramData\BOINC_2 –gui_rpc_port 9999, which i guess tells the second BOINC client to communicate w/ the second BOINC manager over port 9999. (do not break it, as running the second instance of BOINC requires it to be open for port communication).
5) run boinc.exe (which will open to the first BOINC installation by default). in the advanced menu of the BOINC manager, select "launch another BOINC manager." in the data fields that appears, select localhost:9999 (or enter it if it doesn't appear in the drop down box), and enter the password. if the "attach to project" window/wizard appears, then everything is working and ready to go.


...well i'm having a problem right at the end of step 5. when i launch a second BOINC manager, select "localhost:9999," and enter my password, i get the following error:

i don't understand exactly what that means, but note the two separate instances of BOINC clients running in the system tray, and although you can't see it in the screen shot, there are two BOINC managers running...its just that one is directly overlaying the other. at any rate, i don't know how many of you have experience with trying to do this, but any help would be appreciated.

thanks,
Eric

PS - just in case anyone was wondering why i feel its necessary, i'm having some minor side effects to running separate projects on separate GPUs on the same host. to be specific, i like to run Milkyway@Home on my HD 6950 and POEM@Home on my HD 5870. if i run one project or the other, i have no problem maintaining a work buffer. but when both projects are running simultaneously, both projects will completely deplete their work buffers before downloading another chunk of tasks. while its annoying, that reason for wanting to change it only secondary. the primary reason is b/c i believe it to be limiting my efficiency to the tune of 50K-100K PPD.

 

[salvorhardin]

I had to use 127.0.1:9999 in order to use 2 boinc managers. I also had problems maintaining a queue when doing 2 projects with a single manager. Basically it was maintaining my work queue between both projects combined instead of each project separate.

 

[Sunny129]

you are exactly the person i was hoping would respond! i knew there was another thread that talked about this, and that both you and i had posted in it. i guess i fail at searching lol, b/c i really tried to not start a new thread. at any rate, your suggestion solved the problem...now i just have to set up each client to ignore the proper GPUs and run MW@H and POEM@H on separate clients.

by the way, how did you know to try 127.0.0.1:9999 after localhost:9999 didn't work for you?

 

[salvorhardin]

I kept trying different combinations and they kept failing. Then it dawned on me that localhost was referring to the same computer and remembered that 127.0.0.1 refers to the computer. Years ago I would read about people putting 127.0.0.1 in a windows file (don't remember the name) to block out advertising from certain ip addresses. It would show blanks because it would refer to the same computer.

 

[Sunny129]

well i was certain that running MW@H and POEM@H on separate clients would solve my work buffer issues, but it didn't. its like each client is aware of what the other is doing, b/c the work buffers for both projects are being run dry every time before replenishing themselves...despite the projects being run on separate clients. i don't get it...then again, there are alot of things i don't get (and can't stand) about the BOINC infrastructure. maybe one of these days we'll be able to run multiple GPU projects on multiple GPUs on the same host all under a single BOINC client without having to deal with the downside of the inherent flaws in the algorithms that control work buffer functionality.

 

[salvorhardin]

I was having the same problems when running poem and einstein about 2 months ago. I would get 2 tasks from poem and I wouldn't get anymore until I would update the project (even then I would only get 2 wus). Surprisingly only when using an appinfo in poem, If I ran poem without an appinfo I would always have a buffer. I made backups of all my appinfo and ccconfig files and completely uninstalled boinc (program, data directories everything boinc related). Redid all my directories and it seemed to fix my buffer problems.

 

[Sunny129]

yeah...my host downloads a max of 33 POEM@Home tasks at a time, which isn't the worst thing in the world b/c its enough to keep my HD 5870 busy for ~6.5 hours running 6 of them in parallel. my major complaint with this actually isn't the moderate possibility of delayed communications w/ the server, but rather the strong possibility that my host will end up crunching less than 6 in parallel as it crunches through the last few tasks in the queue (for instance, if i get the max of 33 tasks from the server, my GPU will crunch through the first 30 tasks, and will then have to crunch the remaining 3 tasks in parallel before it'll allow me to download more work. my GPU is far more efficient crunching 6 tasks in parallel than it is crunching 3 tasks in parallel...and sometimes the scenario is even worse. sometimes i'll get 19 tasks from the server for instance, in which case my GPU will obviously crunch through the first 18 tasks in groups of 6, and then have to crunch the last remaining task by itself before BOINC will fetch more work.

my major complaint w/ Milkyway@Home is the possibility, or should i say certainty, of decreased compute efficiency. you see, it is a common misconception that running 2 MW@H tasks simultaneously isn't anymore efficient than running 1 at a time b/c if i run a pair of MW@H tasks, they seem to take approx. twice the amount of time it takes the same GPU to crunch 1 task. however, there is a 3 to 10-second window (depending on the CPU and GPU being used) just after a task reaches 100%, during which the entire workload is transferred to the CPU and GPU utilization drops to zero. and so if i take a group of MW@H tasks, start one task, and then start a 2nd task when the first task reaches ~50% progress, i can see that when the first task reaches 100% and all work associated with it transfers to the CPU, the 2nd task (which is now probably near the 50% mark) keeps the GPU loaded. now i know 4 seconds of GPU downtime doesn't sound like much, but when it happens every 60-120 seconds, it adds up. in the worst case scenario (i.e. if i were running only 1 MW@H task at a time), this would happen on my host 1394 times a day. multiply that by 4 seconds, and that's 5575 seconds, or 1:33:00. that's alot of downtime in a day. in the least worst case scenraio (i.e. when i'm running 2 MW@H tasks in parallel, but the start of 2 tasks running at any given time won't stay offset for any appreciable amount of time due to the work fetch issues caused by running MW@H alongside POEM@Home), i'll have 46.5 min. of GPU idle time in a day. obviously if i suspend POEM@Home and only run MW@H, my work fetch issues go away, and therefore the buffer stays full, and therefore tasks continue to start (and finish) offset from each other and won't allow the GPU to see any idle time throughout the day.

i suppose i really should have called it a work fetch issue from the beginning, instead of a work buffer issue. i mean yes, not being able to keep a decent work buffer for either of my GPU projects is most definitely a problem...but its being caused by the work fetch issues that BOINC is creating b/c it doesn't think i can manage my own work buffers while running 2 separate projects on 2 separate GPUs all in the same host. perhaps i'll try what you did and back up all that BOINC data (and perhaps even the entire BOINC data directory), but that seems like a pain in the @ss, and i don't have the patience for it tonight anyway.

by the way, upon putting things back to the way they were, i kept seeing multiple instances of boinc.exe (the client) in the windows task manager, and MW@H tasks were running ridiculously slow. i finally remembered that i had forgotten to remove the <allow_multiple_clients>1</allow_multiple_clients> line from my cc_config.xml file. once i did that, everything went back to normal...i suppose i could have just changed the 1 to a 0 as well. i know that's really neither here nor there, but i just thought i'd mention it in case someone has problems with the same thing or something similar...

 

[CanoeBeyond]

PS - just in case anyone was wondering why i feel its necessary, i'm having some minor side effects to running separate projects on separate GPUs on the same host. to be specific, i like to run Milkyway@Home on my HD 6950 and POEM@Home on my HD 5870. if i run one project or the other, i have no problem maintaining a work buffer. but when both projects are running simultaneously, both projects will completely deplete their work buffers before downloading another chunk of tasks. while its annoying, that reason for wanting to change it only secondary. the primary reason is b/c i believe it to be limiting my efficiency to the tune of 50K-100K PPD.

Hi Eric,

It appears this is a BOINC bug that occurs when using:

1) a BOINC version newer than 7.0.2 and 2 or more ATI GPUs
2) and project exclusions in the cc_config
3) and app_info files in the project folders.

I just posted a bug report:

http://boinc.berkeley.edu/dev/forum_thread.php?id=7796

More info in this thread:

http://boinc.fzk.de/poem/forum_thread.php?id=786&nowrap=true#6890

Hope this helps. It was a real ah heck to track down. Not sure if it occurs when using multiple NVIDIA GPUs with exclusions and app_infos as I have no dual NVIDIA boxes.

And GREETINGS from the crunchers at Ars

Edit: BTW, if you care to include your experience on the above threads please do so as with more reports we just may get this bug fixed faster.

 

[Sunny129]

Hi Eric,

It appears this is a BOINC bug that occurs when using:

1) a BOINC version newer than 7.0.2 and 2 or more ATI GPUs
2) and project exclusions in the cc_config
3) and app_info files in the project folders.

I just posted a bug report:

http://boinc.berkeley.edu/dev/forum_thread.php?id=7796

More info in this thread:

http://boinc.fzk.de/poem/forum_thread.php?id=786&nowrap=true#6890

Hope this helps. It was a real ah heck to track down. Not sure if it occurs when using multiple NVIDIA GPUs with exclusions and app_infos as I have no dual NVIDIA boxes.

And GREETINGS from the crunchers at Ars

Edit: BTW, if you care to include your experience on the above threads please do so as with more reports we just may get this bug fixed faster.

hey what's up Beyond! cool that you found my thread here. i take it you noticed that, aside from my dual ATI/AMD GPU box which you obviously know has the same troubles as yours, one of my other boxes has dual GTX 560 Ti's. i'm currently participating in the SETI@Home Wow! event until August 30th or so. but i could test for this issue once that race ends. typically they both run Einstein@Home BRP4 tasks, so i'm not using any exclude statements, and therefore don't have the work fetch problem on that machine. i also don't use an Einstein@Home app_info.xml b/c i can run multiple tasks in parallel via my web preferences. but i suppose the app_inf.xml in my SETI@Home folder and the exclude statements should be enough to set things off once i start running both E@H and S@H.

btw, welcome to the AnandTech forums

 

[CanoeBeyond]

hey what's up Beyond! cool that you found my thread here. i take it you noticed that, aside from my dual ATI/AMD GPU box which you obviously know has the same troubles as yours, one of my other boxes has dual GTX 560 Ti's. i'm currently participating in the SETI@Home Wow! event until August 30th or so. but i could test for this issue once that race ends. typically they both run Einstein@Home BRP4 tasks, so i'm not using any exclude statements, and therefore don't have the work fetch problem on that machine. i also don't use an Einstein@Home app_info.xml b/c i can run multiple tasks in parallel via my web preferences. but i suppose the app_inf.xml in my SETI@Home folder and the exclude statements should be enough to set things off once i start running both E@H and S@H.

btw, welcome to the AnandTech forums

Thanks! I found the thread when googling for problems similar to what I was experiencing. If you can do a test with the NVidias when you get a chance it would be great. More information for the bug fixers.

I appreciate it. Drop in on the Ars forum any time and have a great weekend!

 

[StitchExperimen]

I was running WCG and it doesn't use all the clock cycles when set at 100% so I had a virtual machine running Ubintu LTR and loaded up Boinc and ran another project.
My problem which I didn't pursue was getting a GPU to run in a virtual machine which was VMware Player at the time. Also to get it to connect on the network I had to change the NIC to a bridge setting.
If you haven't worked with a VM before a hint is when it says to press CTRL-ALT-DEL to log in you do that by going into the tabs at the top of the VM's settings and click on send CTRL-ALT-DEL.
Have a nice day.

 

[Sunny129]

Thanks! I found the thread when googling for problems similar to what I was experiencing. If you can do a test with the NVidias when you get a chance it would be great. More information for the bug fixers.

I appreciate it. Drop in on the Ars forum any time and have a great weekend!

hey Beyond, i had a chance to test for the strange download behavior on my dual nVidia GTX 560 Ti box, and it does persist. i posted a detailed description of the tests i ran and their results in your Strange ATI WU download behavior thread on the POEM message boards.

 

[commhealy]

So I must be doing something wrong. In the very first command: &#8220;c:\Program Files\BOINC\boinc.exe&#8221; &#8211;allow_multiple_clients &#8211;dir g:\ProgramData\BOINC_2 (i only edited the drive letters) I get this message : THe filename, directory name, or volume label is incorrect.
Any ideas? BOINC is closed and I created the required directory