ntpMerlin What is ntpMerlin?

[heysoundude]

This is a great and informative video. Thanks for sharing.

a Pleasure.
now get the right MTU/MRU for your LAN <--> ISP connection and with QoS, you'll be just about as good as it can get. It's not a pipe, it's a bucket brigade.

 

[SilentStorm]

a Pleasure.
now get the right MTU/MRU for your LAN <--> ISP connection and with QoS, you'll be just about as good as it can get. It's not a pipe, it's a bucket brigade.

What's the best way to do that?

 

[heysoundude]

What's the best way to do that?

I'll see if I can find the links I posted...probably in one of Cake-QoS topics/threads...pre-Merlin integration, v384.18 IIRC...standby.

 

[pdc]

That was a really cool video, watching all those metronomes in sync was downright spooky! Speaking of oscillations, check this out:

I'm pretty sure it is an analog to ambient temperature. Though also you know how Merlin 386 increases CPU temp? Guess when I installed it? (The second spike was one of the betas). I'm not positive that's why the oscillation, since the low values aren't quite like clockwork, but it is a 24-hour cycle. And this time of year the ambient is controlled by the furnace thermostat, so pretty regular. We'll see now that spring is here...

 

[Wade Coxon]

I'm just going to leave this right here for your (and everyone else's) consideration:

It's a neat video, but is not applicable to the topic of ntp. ntp does not synchronise the clock pulses or "ticks" between different devices, or even synchronise them within a single device. It just adjusts the device's time-of-day to compensate for imprecise "ticking".

 

[L&LD]

@Wade Coxon, and isn't that the very definition of synchronization?

That video shows that we don't think we know everything we think we know. Until we know that 'one more thing'.

If chemical synchronization can occur, I don't see why it can't happen with electrical systems. Particularly ones that are tied together.

 

[Wade Coxon]

@Wade Coxon, and isn't that the very definition of synchronization?

That video shows that we don't think we know everything we think we know. Until we know that 'one more thing'.

ntp adjusts the router's time-of-day clock to remain more-or-less in-step with the time-of-day clock of an accurate time source. That's it, that's the limit of the synchronisation that is taking place. ntp does nothing to improve the router's underlying oscillator drift (it just measures it), or synchronise the inter-operation of the router's underlying silicon.

Having an time-of-day clock does not improve your router's networking/routing performance, that is the only fact I am trying to explain here.

 

[L&LD]

I may not be able to explain it (logically or not), but using chronyd makes the network, um, 'smoother' and 'more direct'. Not faster, but more instant. The same kind of effect the use of an RT-AX86U has on a network too. Not smart enough to even describe it accurately. But smart enough to notice it though.

 

[Wade Coxon]

I may not be able to explain it (logically or not)

Placebo effect unfortunately.

 

[Jack Yaz]

Honestly with all the misinformation about ntp and network performance flying around lately I'm very tempted to discontinue ntpmerlin publicly and just use it myself.... I'm getting rather irritated by it all being on threads with the ntpmerlin prefix.

 

[MvW]

Honestly with all the misinformation about ntp and network performance flying around lately I'm very tempted to discontinue ntpmerlin publicly and just use it myself.... I'm getting rather irritated by it all being on threads with the ntpmerlin prefix.

No, please don't. To keep my clock ticking I need those sexy graphs
Please don't take that away from me

 

[L&LD]

@Jack Yaz, please don't take what I post about the chronyd network effects as 'misinformation'. There have been at least a few 'lay' customers who have asked me what I did that made their network so responsive (after a week and a month +) of my installing ntpMerlin and enabling chronyd. I never told them anything I've repeated here (above). They came to that conclusion totally on their own. The network has less latency, and it is noticeable by many different people if not all.

Is there some other logical explanation? Yes, it is possible. I just know the results are seen when chronyd (via ntpMerlin, I don't know any other way to install/enable this on a router) is introduced.

 

[Serkan K.]

I'm just going to leave this right here for your (and everyone else's) consideration:

Thank you for the video. I was amazed during the clapping crowd scene thinking how many times I've experienced that and never ever thought of syncing. I always thought that it was the wish for an encore.

 

[heysoundude]

@Jack Yaz is the 3rd carrier of this ball since 2015 by my account https://www.snbforums.com/threads/ntp-daemon-for-asuswrt-merlin.28041/ and I've been looking for posts I made back then with links that explain it better than it seems I'm currently able to, but I've failed.

I'll take credit/blame for being the sh!t disturber here. we're talking absolute and relative times, and ntpMerlin relays an absolute time reference to clients for them to get in line with by adjusting their relative time to. I'm not saying crystals on various mobos sync like the metronomes in that video (they might - I'm not sure someone has ever looked. if the routers ran on AC, they would probably follow the power frequency or a multiple of it), I'm saying the various operating systems/devices look to ntp for their own "I'm early/late" internal check and pull themselves into lockstep with others on the network

In my day job, with a bunch of unrelated clocks freewheeling to no absolute time reference - or to the one they determine to be their own - bad things happen (see an earlier reference to a video/television example w/r/t my handle here - eyes seeing lips not matching what ears are hearing them say - and that's just the tip of the iceberg: piccolos can be made to sound like stuttering clarinets, or cellos like violins, or sopranos like baritones when my world isn't syncing). When all of those outlier clocks follow a master, the systems become cohesive and those problems cease to exist/occur.

It works. Router follows timesource and broadcasts it locally for devices on LAN to line up to/with. small nudges to those clocks keep everything flowing without needing significant correction, and thats what smooths or makes things seem snappier or "faster" after our routers "settle in"

ntpMerlin surveys various external time sources designed and proven to be ridiculously accurate, adjusts itself to be in line with the one that it has the tightest communication with, and broadcasts that reference locally for the devices on the network to follow... @L&LD 's clients are seeing/feeling/experiencing this, because suddenly the devices on their networks arent having to slew and pull themselves up/down/allroundthetown as far to fall in line with each other to line up with the bucket brigade of data flowing into, out of and within the network as established by the router.

I'ver agonized long enough over posting this reply. shag it.

 

[archiel]

Can somebody assist. I am trying to find a way to record the ntp connections
I can use watch with conntrack to give me a snapshot e.g. watch -tn10 'cat /proc/net/nf_conntrack | grep port=123' and this gives results such as

ipv4     2 udp      17 25 src=90.213.99.16 dst=194.80.204.184 sport=41025 dport=123 src=194.80.204.184 dst=90.213.99.16 sport=123 dport=41025 mark=0 iqprio=0 swaccel=0 hwaccel=0 use=2

, however this is overwritten on the next update. What I want is to see the cumulative update, is there a way to do this using watch or other tools?

I am doing this so that I can better understand what is happening with intercepted traffic under ipv4 and then extend that to try to see if what happens when I add ipv6 into the mix.
[the source ip has been edited, it is not the router's real wan ip]

 

[ColinTaylor]

You could use tcpdump, but if you're specifically interested in the conntrack entries:

# conntrack -E | grep "dport=123"
    [NEW] udp      17 30 src=192.168.1.49 dst=51.145.123.29 sport=123 dport=123 [UNREPLIED] src=51.145.123.29 dst=80.7.99.99 sport=123 dport=123
 [UPDATE] udp      17 30 src=192.168.1.49 dst=51.145.123.29 sport=123 dport=123 src=51.145.123.29 dst=80.7.99.99 sport=123 dport=123
    [NEW] udp      17 30 src=192.168.1.10 dst=91.189.94.4 sport=32899 dport=123 [UNREPLIED] src=91.189.94.4 dst=80.7.99.99 sport=123 dport=32899
 [UPDATE] udp      17 30 src=192.168.1.10 dst=91.189.94.4 sport=32899 dport=123 src=91.189.94.4 dst=80.7.99.99 sport=123 dport=32899
[DESTROY] udp      17 src=192.168.1.49 dst=51.145.123.29 sport=123 dport=123 src=51.145.123.29 dst=80.7.99.99 sport=123 dport=123
[DESTROY] udp      17 src=192.168.1.10 dst=91.189.94.4 sport=32899 dport=123 src=91.189.94.4 dst=80.7.99.99 sport=123 dport=32899

Notice how the Windows 7 client at 192.168.1.49 uses a source port of 123 whereas the Ubuntu client at 192.168.1.10 uses an ephemeral source port.

 

[archiel]

You could use tcpdump, but if you're specifically interested in the conntrack entries:

# conntrack -E | grep "dport=123"
    [NEW] udp      17 30 src=192.168.1.49 dst=51.145.123.29 sport=123 dport=123 [UNREPLIED] src=51.145.123.29 dst=80.7.99.99 sport=123 dport=123
[UPDATE] udp      17 30 src=192.168.1.49 dst=51.145.123.29 sport=123 dport=123 src=51.145.123.29 dst=80.7.99.99 sport=123 dport=123
    [NEW] udp      17 30 src=192.168.1.10 dst=91.189.94.4 sport=32899 dport=123 [UNREPLIED] src=91.189.94.4 dst=80.7.99.99 sport=123 dport=32899
[UPDATE] udp      17 30 src=192.168.1.10 dst=91.189.94.4 sport=32899 dport=123 src=91.189.94.4 dst=80.7.99.99 sport=123 dport=32899
[DESTROY] udp      17 src=192.168.1.49 dst=51.145.123.29 sport=123 dport=123 src=51.145.123.29 dst=80.7.99.99 sport=123 dport=123
[DESTROY] udp      17 src=192.168.1.10 dst=91.189.94.4 sport=32899 dport=123 src=91.189.94.4 dst=80.7.99.99 sport=123 dport=32899

Notice how the Windows 7 client at 192.168.1.49 uses a source port of 123 whereas the Ubuntu client at 192.168.1.10 uses an ephemeral source port.

Thank for this, I was (am) definitely interested in the conntrack entries, as when ntpmerlin is in intercept mode, tcpdump will report the intended connection whereas conntrack will show the effect of the prerouting command (at least that is how I understand it)

For example i have one of my windows pcs setup to use 17.253.34.253 as its time source, if I run tcpdump -nn port 123 -i any then I get

17:15:09.721031 IP 10.0.10.116.123 > 17.253.34.253.123: NTPv3, Client, length 48
17:15:09.721031 IP 10.0.10.116.123 > 17.253.34.253.123: NTPv3, Client, length 48
17:15:09.721212 IP 17.253.34.253.123 > 10.0.10.116.123: NTPv3, Server, length 48
17:15:09.721218 IP 17.253.34.253.123 > 10.0.10.116.123: NTPv3, Server, length 48

- no sign that the ntp traffic is being handled on the router, but if i am looking at conntrack I (think I) see the intercept

ipv4     2 udp      17 16 src=10.0.10.116 dst=17.253.34.253 sport=123 dport=123 src=10.0.10.1 dst=10.0.10.116 sport=123 dport=123 mark=0 iqprio=0 swaccel=0 hwaccel=0 use=2

 

[BreakingDad]

It's just... those sexy graphs which make me feeling all warm and fuzzy

Glad it's not just me.

 

[MvW]

@Jack Yaz, it looks like you're outnumbered... So those sexy graphs 'ain't goin' nowhere

 

[Jack Yaz]

@Jack Yaz, it looks like you're outnumbered... So those sexy graphs 'ain't goin' nowhere

unfortunately there's an electoral college and i am in a "state" with a controlling interest