In principle, one could take a marginal GPU or system over the edge by increasing the loading, but doing so using the project preference setting is no more dangerous, in my opinion, than doing so with a configuration file.
Quote:
the effects of a web based adjustments on GPU's of different power and capacities.
You have four distinct locations (venues) available, which if you are not already using them to other purpose might usefully allow you to collect sufficiently similar hosts together for this and other purposes. On the other hand, if you operate a large and diverse fleet, or have already dedicated this distinction to other uses important to you, that would be a good reason to use the config file method.
Quote:
description or reference for the meanings of BRP, FGRP, and GW
At the moment the project offers only two types of work for graphics cards (and one other for the onchip Intel graphics),
Binary Radio Pulsar Search (Arecibo, GPU)
Binary Radio Pulsar Search (Perseus Arm Survey)
As both are forms of Binary radio pulsar search, both have their GPU multiplicity controlled by the same preference item on the web page GPU utilization factor of BRP apps
FGRP is a Gamma Ray Pulsar search, and GW a Gravitational Wave search. So far as I know there is not at the moment a GPU work offering in either of these categories.
Great response ... and useful ... but I will need to spend some time both with your responses and some testing. Wouldn't classify my machines as a fleet but I do have 6 that I like to keep busy, two of which contain dual GPU's.
So I'm off to some testing. Thanks for the help. I'll get back either with results or questions, hopefully the former.
First:
please place the working buffer in Boincmanager to zero.
Now delete Wu until no High Priority mode will disappear.
With the core reducing the high priority can not be changed, only the correct operation of the app_config for the GPU is enabled.
Trying to follow your directions above to Merle above concerning removal of High Priority wu's. A couple questions -
1. For working buffer, are you referring to "disk and memory usage" parameters in Tools->Preferences?
2. Assuming the buffer is set to 0, should I expect that inactive wu's already in the queue will be assigned high priority when I discard the active high priority wu's or will those in the queue disappear because the buffer is now 0?
2. Assuming the buffer is set to 0, should I expect that inactive wu's already in the queue will be assigned high priority when I discard the active high priority wu's or will those in the queue disappear because the buffer is now 0?
Preferences on work fetch affect new downloads after they are changed, not the handling of work already downloaded.
If for some reason you are bent on making High Priority mode disappear, I have personally seen it go away after I have suspended a sufficient number of WUs on the machine.
I should hasten to add that if you have any units suspended, your host will not request new work of any type to download. As this effect goes away when there are not longer any units in suspend, you can get a gusher if you have not somehow reduced the download request that got you to High Priority in the first place.
If for some reason you are bent on making High Priority mode disappear, I have personally seen it go away after I have suspended a sufficient number of WUs on the machine.
Actually, I wanted to attempt adjusting the web parameters first but needed to be prepared for a fallback should I stall the slower machine with overload. Also, I don't have a grasp of your comments about the 4 venues you mentioned. Four independently controlled paths would likely meet my needs since 2 of my 6 machines do not have GPU's. Any leads you have there would be helpful.
If for some reason you are bent on making High Priority mode disappear, I have personally seen it go away after I have suspended a sufficient number of WUs on the machine.
Actually, I wanted to attempt adjusting the web parameters first but needed to be prepared for a fallback should I stall the slower machine with overload. Also, I don't have a grasp of your comments about the 4 venues you mentioned. Four independently controlled paths would likely meet my needs since 2 of my 6 machines do not have GPU's. Any leads you have there would be helpful.
As always, appreciate your response.
Go to Your Account on the webpage, click Preferences for this project Einstein@Home preferences and under it you will see the different 'venues'...default, work, home and school. You can setup each to crunch any kind of units you want, mixing and matching however you like. When you are done go back into Your Account and click on Computers on this account View. Click on Details on each pc and at the very bottom of that page you can switch them to the 'venue' of your choice, once you then do an update on each pc, either manually, or you can let it do it automatically, the changes will take affect for all future workunits.
If for some reason you are bent on making High Priority mode disappear, I have personally seen it go away after I have suspended a sufficient number of WUs on the machine.
Actually, I wanted to attempt adjusting the web parameters first but needed to be prepared for a fallback should I stall the slower machine with overload.
Just to make sure you understand the terminology:
High priority in Boinc = Run tasks in "earliest deadline first" mode. It's to try and get the work back before the deadline is up and has nothing to do with process priority in the OS, so there should be no other effect than the tasks in the cache getting run in a different order than "first in, first out" aka FIFO.
If you set your prefs through Boinc manager these will always override the web based prefs. Only "Computing prefs" can be set in Boinc manager, project prefs and community prefs must be set via the web page. If you want to revert to web based prefs then open the prefs settings in Boinc manager and click the "Clear" button.
If for some reason you are bent on making High Priority mode disappear, I have personally seen it go away after I have suspended a sufficient number of WUs on the machine.
Actually, I wanted to attempt adjusting the web parameters first but needed to be prepared for a fallback should I stall the slower machine with overload. Also, I don't have a grasp of your comments about the 4 venues you mentioned. Four independently controlled paths would likely meet my needs since 2 of my 6 machines do not have GPU's. Any leads you have there would be helpful.
As always, appreciate your response.
Go to Your Account on the webpage, click Preferences for this project Einstein@Home preferences and under it you will see the different 'venues'...default, work, home and school. You can setup each to crunch any kind of units you want, mixing and matching however you like. When you are done go back into Your Account and click on Computers on this account View. Click on Details on each pc and at the very bottom of that page you can switch them to the 'venue' of your choice, once you then do an update on each pc, either manually, or you can let it do it automatically, the changes will take affect for all future workunits.
When I read something like this, the though comes to mind, "Why didn't I think of this?" Disappointed that I didn't but surely appreciate the pointer.
If for some reason you are bent on making High Priority mode disappear, I have personally seen it go away after I have suspended a sufficient number of WUs on the machine.
Actually, I wanted to attempt adjusting the web parameters first but needed to be prepared for a fallback should I stall the slower machine with overload.
Just to make sure you understand the terminology:
High priority in Boinc = Run tasks in "earliest deadline first" mode. It's to try and get the work back before the deadline is up and has nothing to do with process priority in the OS, so there should be no other effect than the tasks in the cache getting run in a different order than "first in, first out" aka FIFO.
If you set your prefs through Boinc manager these will always override the web based prefs. Only "Computing prefs" can be set in Boinc manager, project prefs and community prefs must be set via the web page. If you want to revert to web based prefs then open the prefs settings in Boinc manager and click the "Clear" button.
Appreciate the clarification. Had thought that it was Einstein's attempt in Boinc to seize compute power away from other projects. Your comments about preference priorities will help me tune up for Einstein.
Well I now see that BRP is
)
Well I now see that BRP is Binary Radio Pulsar and GW is Gravitational Wave. Still unable to locate the FGRP.
RE: one being the
)
In principle, one could take a marginal GPU or system over the edge by increasing the loading, but doing so using the project preference setting is no more dangerous, in my opinion, than doing so with a configuration file.
You have four distinct locations (venues) available, which if you are not already using them to other purpose might usefully allow you to collect sufficiently similar hosts together for this and other purposes. On the other hand, if you operate a large and diverse fleet, or have already dedicated this distinction to other uses important to you, that would be a good reason to use the config file method.
At the moment the project offers only two types of work for graphics cards (and one other for the onchip Intel graphics),
As both are forms of Binary radio pulsar search, both have their GPU multiplicity controlled by the same preference item on the web page
GPU utilization factor of BRP appsFGRP is a Gamma Ray Pulsar search, and GW a Gravitational Wave search. So far as I know there is not at the moment a GPU work offering in either of these categories.
Great response ... and useful
)
Great response ... and useful ... but I will need to spend some time both with your responses and some testing. Wouldn't classify my machines as a fleet but I do have 6 that I like to keep busy, two of which contain dual GPU's.
So I'm off to some testing. Thanks for the help. I'll get back either with results or questions, hopefully the former.
RE: First: please place
)
Trying to follow your directions above to Merle above concerning removal of High Priority wu's. A couple questions -
1. For working buffer, are you referring to "disk and memory usage" parameters in Tools->Preferences?
2. Assuming the buffer is set to 0, should I expect that inactive wu's already in the queue will be assigned high priority when I discard the active high priority wu's or will those in the queue disappear because the buffer is now 0?
Appreciate your advice ...
RE: 2. Assuming the
)
Preferences on work fetch affect new downloads after they are changed, not the handling of work already downloaded.
If for some reason you are bent on making High Priority mode disappear, I have personally seen it go away after I have suspended a sufficient number of WUs on the machine.
I should hasten to add that if you have any units suspended, your host will not request new work of any type to download. As this effect goes away when there are not longer any units in suspend, you can get a gusher if you have not somehow reduced the download request that got you to High Priority in the first place.
RE: If for some reason
)
Actually, I wanted to attempt adjusting the web parameters first but needed to be prepared for a fallback should I stall the slower machine with overload. Also, I don't have a grasp of your comments about the 4 venues you mentioned. Four independently controlled paths would likely meet my needs since 2 of my 6 machines do not have GPU's. Any leads you have there would be helpful.
As always, appreciate your response.
RE: RE: If for some
)
Go to Your Account on the webpage, click Preferences for this project Einstein@Home preferences and under it you will see the different 'venues'...default, work, home and school. You can setup each to crunch any kind of units you want, mixing and matching however you like. When you are done go back into Your Account and click on Computers on this account View. Click on Details on each pc and at the very bottom of that page you can switch them to the 'venue' of your choice, once you then do an update on each pc, either manually, or you can let it do it automatically, the changes will take affect for all future workunits.
RE: RE: If for some
)
Just to make sure you understand the terminology:
High priority in Boinc = Run tasks in "earliest deadline first" mode. It's to try and get the work back before the deadline is up and has nothing to do with process priority in the OS, so there should be no other effect than the tasks in the cache getting run in a different order than "first in, first out" aka FIFO.
If you set your prefs through Boinc manager these will always override the web based prefs. Only "Computing prefs" can be set in Boinc manager, project prefs and community prefs must be set via the web page. If you want to revert to web based prefs then open the prefs settings in Boinc manager and click the "Clear" button.
RE: RE: RE: If for
)
When I read something like this, the though comes to mind, "Why didn't I think of this?" Disappointed that I didn't but surely appreciate the pointer.
RE: RE: RE: If for some
)
Appreciate the clarification. Had thought that it was Einstein's attempt in Boinc to seize compute power away from other projects. Your comments about preference priorities will help me tune up for Einstein.