|
From: | Martin Boissonneault |
Subject: | Re: Clarifications about PPS SHM content |
Date: | Fri, 3 Apr 2020 12:15:15 -0400 |
User-agent: | Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:68.0) Gecko/20100101 Thunderbird/68.6.0 |
Hi all,
Just a quick update: My jitter wasn't caused by ntpheat.
For the past few days, I turned off ntpheat and this bad ~700ns
jitter (had ~265ns before) is continuing. I've been hunting the
cause, the clue I have is a negative clock offset spike every
45min.
So far, I think it's the timeout period of the ADS-B decoders
(~3888s). My current hypothesis is that there are so few planes up
there that the feeders/decoders restart because they haven't seen
a plane in 45min. So, I've shut down the ADS-B decoders/feeders.
If I'm right, I'll know in a few hours! Then, if it's fixed, I'll be able to brag about my NTP server being affected by COVID-19 ;-)
Stay safe,
Martin
Hi!
I ran a test with ntpheat for a day, and my jitter StdDev jumped from
~275ns to ~700ns. For my setup, it's NOT better.
I'll see if my physical "mod" will help: I stuck two layers of sticky
thermal pads on top of each other, covering both the CPU footprint and
the crystal, on the underside of the CPU. My idea is to better conduct
heat from the CPU to the crystal while reducing heat transfer transfer
to air. The pads have better thermal conductivity than air, but being on
top of components like a blanket should provide some isolation to air.
Heat should spread a bit better between nearby components and the PCB,
but I have nothing on top to wick heat away. If I had thicker pads (mine
are pretty thin), I'd try to put a piece of metal to spread heat across
the whole area.
My CPU usage is already 25%x4 cores, there is some heat there. I just
have to stabilize it across the xtal and prevent too much air movement.
I will report back later!
Martin
=========================================
Thanks for your initial report, Martin! I look forward to further tests. In the light of your tests, I've updated my Web page to indicate that it's not a cure-all, rather a try-it-and-see solution.
Cheers,
David
[Prev in Thread] | Current Thread | [Next in Thread] |