AVS-10/4 Underclocking

[bennunemaker]

I received my AVS-10/4 and it is running very well.  I am happy with the low power usage.

 

I would like to see if I can take things a little further, though.  I upgraded the system to the more efficient power supply, which undoubtedly helps save a few additional watts.

 

1. I would like to underclock the CPU a little bit, what is the process to do this?  The system is headless.

2. Can I further slow down the fans a little bit?  It is much quieter than my last server but actually still louder than I was hoping.

3. Does anyone have any other suggestions for neat uses of this or ways to save more power/noise?  I am using it for all my data in addition to Time Machine -- nice to not have to buy a Time Capsule.

 

This server has 5 * 4TB drives + 1 * 2TB drive (Time Machine and Misc Backups).  I already modified Time Machine to only backup once every 24-hours because I don't like the backup drive spinning up all the time.

 

An annoyance I have is that when I'm streaming data from my share which happens to be on disk1, if disk2 spins up, the stream is halted momentarily while the drive spins up.  Perhaps this is a limitation of the network file system transport meaning that these requests are effectively serialized and blocking?

[mr-hexen]

re: the spinup causing stream interruption. put the drives on the same spinup group.

 

this way when a movie starts playing from disk1 it will spin up all the drives at that time and keep them spinning until disk1 can sleep again.

[garycase]

... I upgraded the system to the more efficient power supply, which undoubtedly helps save a few additional watts.

 

While it may SEEM like using an 80+ Gold unit is more efficient than using an 80+ Bronze unit, that's not necessarily the case ... and in fact is almost certainly NOT the case with regards to this server.

There's a 5% difference in the rated efficiency of Bronze vs. Gold units (82% vs. 87%  or  85% vs. 90%, depending on the load point being tested) ==> HOWEVER, neither of these certifications include any requirements for efficiency at loads below 20%.  It's not uncommon for PSU efficiency to drop significantly -- 10% to 15% -- at lower loads.    Note that the 520w Bronze unit is still in its 80+ certified range at a 104w load, whereas the 650w unit's range starts 130w.    So at lower loads ... which are VERY likely with this server [in fact, I suspect it NEVER hits the 80+ range of the 650w unit] ... the efficiency of BOTH of the available power supplies will be lower than their 80+ certifications.

 

Whether or not a 5% "head start" in efficiency offsets the greater efficiency drop due to a lower load percentage is hard to say.  You'd really have to have identically configured units except for the PSU's and do some measurements to know for sure => but I'd except a "typical" power draw of the Atom-based unit to be in the 50w range, and the i3 unit to be perhaps 10-15w higher.  In fact, even when running parity checks with a full complement of drives, it's unlikely either unit ever gets above the minimum 20% power draw of the 650w unit ... so it's always below the 80+ certified range.

 

If I was building an identical server and wanted a more efficient power supply than the 520w 80+ Bronze unit, the "upgrade" I'd make would be to a SMALLER 80+ Gold unit, like the excellent Seasonic 450w unit (which will easily support a full complement of drives as long as you're using low-power WD Reds or Seagate NAS drives or either's "green" series drives).

 

 

1. I would like to underclock the CPU a little bit, what is the process to do this?  The system is headless.

2. Can I further slow down the fans a little bit?  It is much quieter than my last server but actually still louder than I was hoping.

 

Both of the motherboards LimeTech uses for that server support IPMI, so you can make any changes you want in the BIOS from another system.  However, they're both very efficient boards, and both CPU choices are already very efficient,  so I don't think it's worth bothering.    My 2nd system uses the Supermicro D525 board, and at idle (drives spun down) it only draws 20w ... and I suspect at least a few of those are due to the lower efficiency of the PSU at that low a power draw.    And the Core i3 system probably only draws 10-15 more watts.

 

As for the fans ... you could certainly add a fan controller to the circuit, but I'd be VERY careful about that.  Run a few tests at different rpm levels and see what the drive temps are ==> the best test is to do a parity check and see what the temps are after about an hour.    You do NOT want to slow the fans down so much that the drives run at high temperatures.    Do a "baseline" test => run a parity check with the stock configuration and see what the temps are;  then don't let them get more than 3-5 degrees warmer than that as a tradeoff for lower noise.

 

 

An annoyance I have is that when I'm streaming data from my share which happens to be on disk1, if disk2 spins up, the stream is halted momentarily while the drive spins up.  Perhaps this is a limitation of the network file system transport meaning that these requests are effectively serialized and blocking?

 

It's most likely a limitation of the controller, which "freezes" activity during a spinup of another drive on the same controller.    I don't, however, know why UnRAID would be spinning up another drive if you're sure the entire movie is on the first disk.    Do you have any add-ons that may be using that drive?  ... or some other program on your network that could be accessing it?    UnRAID does have a feature that helps minimize these disruptions -- the "spin-up groups", which let you group drives that are on the same controllers so when one of them is spun up, all of them are.    It can take a bit of experimentation to identify which drives have this interaction issue, but if you get the spin-up groups set correctly, the problem you noted will go away 

[bennunemaker]

Interesting points, makes sense.  I have almost all data on disk1, then 2/3/4 are various videos.  The problem is that different types of videos are on different drives and I really don't want to spin up more drives than are needed.  It looks like the drives are connected to the motherboard in addition to a standalone PCI-e controller, if the drives are on separate controllers does the halt of data reading not occur?  It seems like when I have to spin up disk5, things don't halt so maybe that happens to be on a different controller from everything else.  I would rather only spin up what is being used, especially because the only real alternative would be to just spin them all up at the same time or keep them running 24/7 like I did with my last server--which was great because there was never a spin up delay to wait for.  Naturally, whenever I go to read data there will almost always be a spin up delay because of course everything has spun down.  So I always have to start with an annoying delay, but the power savings is nice.

[garycase]

Spinning up a drive on a different controller isn't likely to cause any delays on streams from drives not attached to that controller.    In fact, it can even be a function of specific channels on the controller ... so, for example, an 8-drive unit may have 2 groups of 4 that interact.  Really hard to say without actually testing them.

 

A fairly simple (but a bit time-consuming) test:

 

=> Spin all drives down

=> Start a stream from one drive, so it spins up

=> While it's playing, copy a file from one of the other drives (obviously causing it to spin up) ... and note whether or not it impacted the stream.

 

... Repeat the above until you've tested all combinations for interaction.    Then just set spin-up groups for those drives that interact  [Or just decide you'll live with the occasional pauses  ]

 

[bennunemaker]

Interestingly enough, I put the drives I have in alternating bays (0,2,4,6, so they would cool better and I only have half the bays filled.  I wonder if this precisely makes the drives use alternating channels that interfere with one another.  I should open up the server and look at the paths, I bet you're right.  I can probably reduce or eliminate this problem without resorting to spin up groups.  I don't mind a delay when I start content, just the fact that it halts other content (music while I spin up a video drive) is annoying to me.

[garycase]

Could very well be that moving the drives could change just how they interact.  As I noted, it's pretty easy conceptually to identify the interactions ... but can definitely take a while.    And you have yet-another variable you can control ... moving the drives to other slots 

 

If you DO choose to do that testing, be sure you keep good notes, so you'll have an "interaction" chart you can reference in the future as you get more drives.    Perhaps a spreadsheet with all slots listed both across and down, with the intersection of slots being a "Yes" or "No" vis-à-vis their spinup interaction.