You are not logged in.
- Topics: Active | Unanswered
#1 2017-11-18 02:35:00
- ghaverla
- Member
- From: Dawson Creek, BC, Canada
- Registered: 2017-06-19
- Posts: 111
UPS, digital power supplies and what jobs use power
I run BOINC jobs (and I am not an expert at this). And I have UPSes which can report on how much power is being used. And for one machine (soon to be 2), I have digital power supplies which can report on how much power is being used at 3.3, 5 and 12V.
At the moment, I am just learning about this, and so I am just doing publicly available BOINC stuff. My intention is to make my own BOINC jobs served by a BOINC server I run. Which allows me the possibility of farming out jobs to other people to work on. Which is partially funny, as at the end of the day, what I am trying to improve is farming in the part of the world where I live.
I have 3 UPS in use (1 not yet in use), and 3 of them are from CyberPower. The outlier is a APC UPS that I have had forever.
NUT (Network UPS Tools) is probably the tool most people should use, unless the UPS they have isn't supported by NUT. Another possible reason to not use NUT, is that the data you want/need is not currently provided by NUT.
So, I am starting from the assumption that NUT is the system installed. And among the things NUT does, is by default it looks for UPS data every 15 seconds.
---
If a computer which has a digital power supply is being "probed", the NUT daemon will be able to provide information about the UPS which is supplying that computer. The power supply itself, can also supply information on what it is supplying. The data provided by the UPS should always be higher than the data supplied by the digital power supply. The difference is the inefficiency of the power supply. All other things (not yet specified) being equal, we are not expecting the difference to be constant. The times when the UPS is providing data are not synchronised and/or sampled at the same times that the digital power supply are providing data. If the UPS in question is providing power to more than one computer, that is another source of difference between the two estimates of how much power is being used.
---
Whether or not any of the computers on our LAN have digital power supplies, we (may) have some information as to how much power is being used.
But, we do have some information as to what processes are using a lot of resources at any given time. And as near as I can tell (and I have a Perl bias), libgtop (a Gnome product?) is the place to start there. And so we want to find the N processes that are using significant resources. Resources could be CPU time, but it may also be disk access or swap. And I am going to guess you only want to note things using more than 10% of the maximum usage. So at any given sampling time, we have the N processes using significant resources in terms of possibly more than one criteria.
---
From lmsensors, we may have data about the temperature of the CPU(s) and/or GPU(s) in use, as well as temperatures for storage devices being used. All of which is an estimate of how much power those items are using.
---
To tie this into BOINC, a person wants:
1. to be able to identify if a process getting significant resources is a BOINC project process.
2. if it is a BOINC project process, what project it comes from.
3. if it is a BOINC project process, is it a CPU only process?
4. if it is a BOINC process split between CPU and GPU, how much power is each using? If multiple GPU are involved, how much is each using?
5. other stuff.
---
None of this is about controlling a UPS. It is about reading data from a UPS. But it does depend on software being present to control a UPS. Or rather, monitor the UPS to determine when to shutdown a system.
---
I'm just starting this. I'm working in Perl (as much as I can).
I am not a security expert, I am seeing security "problems" in NUT. But, I suspect most of this isn't a problem, it is just "theoretical".
What I find as I go on, who knows?
Offline
#2 2017-11-18 04:36:25
Re: UPS, digital power supplies and what jobs use power
As most electrical circuits have some thought in increasing efficiency, their loss/wasted energy usually escapes as heat. So where you see much heat escaping it is not due to severe inefficiency but large amounts of energy going through and some escaping as heat.
So I would presume anything that uses the CPU in extremes for a sustained period is consuming more energy. The the GPU processor must come second when heavy graphics are being processed. Then there should be the ram as we are starting to see ram modules with cooling fins. High speed motors (drives) must consume energy too.
I think people who are into graphic games playing and developing, video editing, GIS systems programmers, must know more about what it takes to melt something. My puzzle comes from android devices, being able to run 4cores and HD graphics in gaming and have no fans or get so hot, while a similar pc needs 3-4 fans to keep from melting.
Offline
