Author Topic: Energy consumption of core  (Read 21092 times)

Matthew

  • Douchebag
  • Addicted
  • *
  • Posts: 567
    • View Profile
Re: Energy consumption of core
« Reply #30 on: January 09, 2008, 10:39:47 pm »
I haven't measured the power usage but it's probably in the 25-40 range during use.

I'd love to see an actual measurement.

tkmedia

  • wants to work for LinuxMCE
  • **
  • Posts: 937
    • View Profile
    • LMCECompatible
Re: Energy consumption of core
« Reply #31 on: January 10, 2008, 12:05:54 am »
Repeat of post this is a better place

Power supplies make a big difference

http://www.80plus.org/

To further reduce power.
I am also playing with
http://www.lesswatts.org/projects/powertop/
and others are now hibernating their MD's
My Setup http://wiki.linuxmce.org/index.php/User:Tkmedia

For LinuxMce compatible  systems and accessories
http://lmcecompatible.com/

Hipper

  • Regular Poster
  • **
  • Posts: 23
    • View Profile
Re: Energy consumption of core
« Reply #32 on: January 10, 2008, 05:22:03 am »
I have been looking for a good power saving solution for my core and I have been leaning towards the Intel e2180. It looks like a few other people have successfuly used them in there cores. I am also planning on using RAID 1 with only two 1TB drives to save a bit more power if I can’t spin the drives down when the core is idle.  Has anyone had issues or success with the e2180 and if so what mobo are you using and how many capture cards (i.e. video encode streams) do you have running at one time?

A first step any development team should take when it comes to power consumption is to minimize CPU and disk usage on polling tasks and use hardware to signal events asynchronously. For example, there is no reason why an update task has to run every few mins (or hours even) when we know that most updates are only available once every few days, weeks or months. Why should a task look every so often for updates or even scan the hard drive? Lets learn from systems like TiVo that simply look once a day at 3am when there should be little activity. Then, once the update process completes the disks should return to the state that they where in prior to the update. If all non user tasks where synchronized to run at a specified time when there is low user load and then return to a sleep state we would not have to worry about drives spinning up when there is no needed user or system load.

Also, I completely agree that LMCE should disconnect its media store from its system store to allow for media drives to spin down. However, it seams that there is a simple solution to this problem, use some form of NAS or another low power system to supply data storage and place that system in sleep mode when not in use (assuming that LMCE does not poll the NAS drives). Also has anyone looked into using 2.5 laptop drives in there core instead of the 3.5 inch high speed (10k rpm) power hogs? I know a lot of enterprise server solutions (specify HP servers) are doing that now, not only for power savings but also for space savings; you can fit a lot more 2.5 inch drives in a thin rack mounted server.

As an intermediate solution would it be possible to run the poling tasks and things like the myth scheduler on an external low power system (e.g. Via’s EPIA systems) and simply bring the core out of a standby or hibernate via wake on LAN (or hardwired power up signal) when specific events occur such as a scheduled recording is coming up or based on a household calendar or when a MD comes online?

Note: I have not yet used LMCE (but I will be soon) thus, I have made some assumptions based on what I have read on the Wiki and in the forums in the last few weeks. That is, I have no first hand experience so please feel free to correct me as needed  ;D

Matthew

  • Douchebag
  • Addicted
  • *
  • Posts: 567
    • View Profile
Re: Energy consumption of core
« Reply #33 on: January 10, 2008, 05:52:11 am »
These big drives come with really fast transfer speed, that LMCE doesn't really use. 750GB drives push something like 375MBps, though the SATA interface gives maybe 300MBps tops, probably more like 250MBps. But the highest bandwidth, video, plays at something like <1.5MBps. If those drives could burst transfer when content is requested from them, they'd have a lot of time to spin down between requests. A whole DVD is only max 4.7GB, so a $65 16GB USB Flash drive could cache 3.4 entire DVDs (probably more than 4 movies). But USB does max 60MBps, though multiple parallel USB (on motherboards supporting truly parallel USB, with full bandwidth for each of maybe 4 USB ports) would argue for 4 $15 GB drives to get 240MBps. Since the movies take hours to play, that could be something like 20-30 seconds bursting the requested movie to Flash, for each hour or two pulling them from the Flash. Those Flash drives can probably also cache the entire LMCE from the disk in just 8-10GB of the 16GB cache. The drives might spend something like 90% of the time spun down, even during heavy use, except when skipping through the actual content of a lot of different titles.

How to structure the storage to put a bank of USB Flash drives in the data path as cache for the hard drives, and set the drives to spin down as part of this power saving profile?

blackoper

  • Veteran
  • ***
  • Posts: 110
    • View Profile
Re: Energy consumption of core
« Reply #34 on: January 11, 2008, 07:40:55 am »
one note about saving power.. I'd also look at using the hdhomerun devices instead of HD capture cards. The power usage is lower and it frees up the pci slots.. not to mention the joy of not having to mess with firmware updates/os upgrade issues

Supposedly that intel 2180 uses less power than the be amd series chips or close to it. I've got a 2180 and with just the mobo and an efficient power supply greater than 85% effiency my power usage was 39 watts. Once you add in pci cards it jumps by about 8 watts per card. Under heavy load the power usage jumps about 15 watts give or take