LAN Jitter?

[Eric_on_Fire]

I started getting significant LAN jitter and occasional dropped packets with my AC86U today. Looking for ideas on what could be the cause before I start flashing firmware and other disruptions. I have a AC86U as router, and another AC86U in AP mode (not AI Mesh), both are running 386.2_2 Merlin. There is a Cisco SG110D-08HP unmanaged PoE switch in the network as well, but removing it has no effect. Normally the router is running Skynet and OpenVPN, disabling both has no effect. All are on ethernet and connections are reported to be 1 Gbps. And, yes, I did the obligatory reboot of all devices.

Continuous ping from router to AP:

Continuous ping from router to a client directly connected (no switch / AP):

Ping directly from a windows client to AP, router, or another client yields similar results:

ping -n 10 192.168.5.87

Pinging 192.168.5.87 with 32 bytes of data:
Reply from 192.168.5.87: bytes=32 time<1ms TTL=128
Reply from 192.168.5.87: bytes=32 time<1ms TTL=128
Reply from 192.168.5.87: bytes=32 time=15ms TTL=128
Reply from 192.168.5.87: bytes=32 time<1ms TTL=128
Reply from 192.168.5.87: bytes=32 time<1ms TTL=128
Reply from 192.168.5.87: bytes=32 time<1ms TTL=128
Reply from 192.168.5.87: bytes=32 time=4ms TTL=128
Reply from 192.168.5.87: bytes=32 time<1ms TTL=128
Reply from 192.168.5.87: bytes=32 time<1ms TTL=128
Reply from 192.168.5.87: bytes=32 time=23ms TTL=128

Ping statistics for 192.168.5.87:
Packets: Sent = 10, Received = 10, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 0ms, Maximum = 23ms, Average = 4ms

 

[Eric_on_Fire]

Re-flashed to Merlin 386.2_2 reset and manually reset everything up and now the within-LAN ping is generally <1ms again with occassional bumps into the single-digits when heavy openvpn traffic is occurring. Watching the ping go up whenever both cores get utilized heavily makes me think I need a quad core router!

 

[heysoundude]

what scripts/add-ons to Merlin are you running?
I believe a QoS scheme suitable to your ISP connection speed may help smooth some things out, and ntpMerlin with chrony should help too. and if your ISP offers native IPv6, you should use it.

 

[Eric_on_Fire]

what scripts/add-ons to Merlin are you running?
I believe a QoS scheme suitable to your ISP connection speed may help smooth some things out, and ntpMerlin with chrony should help too. and if your ISP offers native IPv6, you should use it.

Scripts are SkyNet, SpdMerlin, and ConnMon; the real CPU-hog is OpenVPN client for certain devices. I'm not real concerned about QoS, as ping where I live is going to be crummy, but I rarely fully utilize the internet download bandwidth; upload is another story. My concern is purely within-LAN latency and dropped packets, which have been improved by the reset and manual setup. Here is the ConnMon ping record between AP and Router over past 24-hours. Not bad, although I'd much prefer to consistently see <1ms. I don't think that's in the cards while running OpenVPN on this hardware.

 

[John Fitzgerald]

My concern is purely within-LAN latency and dropped packets

Adjust your MTU (stock 1500):

Ping Test to determine Optimal MTU Size on Router

Use the command prompt to send large packets of data outbound, gradually trying smaller packets until you get a positive reply to determine the optimal MTU size.

kb.netgear.com

How To Find Correct MTU Values on Windows?

The maximum transmission unit (MTU) feature on your router allows you to determine the biggest data size permitted on your connection.

hide.me

How to determine best MTU? - Google Suche

 

[heysoundude]

Scripts are SkyNet, SpdMerlin, and ConnMon; the real CPU-hog is OpenVPN client for certain devices. I'm not real concerned about QoS, as ping where I live is going to be crummy, but I rarely fully utilize the internet download bandwidth; upload is another story. My concern is purely within-LAN latency and dropped packets, which have been improved by the reset and manual setup. Here is the ConnMon ping record between AP and Router over past 24-hours. Not bad, although I'd much prefer to consistently see <1ms. I don't think that's in the cards while running OpenVPN on this hardware.

(I dropped ConnMon in favour of SpdMerlin a while back - I didn't see the point in duplicating statistics/charts. also - zoom out (7d/30d) and switch to logarithmic scale for some perspective - it's not terribly bad)
You've just made the argument for yourself to use QoS - you have an asymmetric connection, where your download capability exceeds the upload significantly: the packets looking to "get out" have to look for spots to step into the traffic. by imposing a "speed limit" and "spacing rule" (2 car lengths kind of thing) you're going to see the consistency you're looking for, or much closer to it. (And yeah, MTU factors in to the equation not insignificantly.)

Could you post which hardware and firmware versions you're using please? Also - have you set up a swap on your entware drive? that'll help too, along with ntpMerlin

 

[itpp20]

Ymmv, but I had something similar and that turned out to be a faulty LAN cable connector which got 'touched' when I changed a switch nearby.

 

[Eric_on_Fire]

I left MTU at 1500 bytes as pings (direct from router or other device through cable modem) are fragmented above 1472.

Hardware is a pair of RT-AC86U with Merlin 386.2_2.

I run ConnMon for within LAN health monitoring and SpdMerlin for internet connectivity health monitoring. In that application, I don't think they are really monitoring the same thing.

I have a 2gb swap file for Entware setup through AMTM on a USB 3 drive.

 

[heysoundude]

If your connection speed is below about 300Mbps down, you could enable CakeQoS-Merlin and see what happens.
(don't get hung up on what spdMerlin reports; ignore what the charts report tell you for a while - it's user experience/network "feel" improvements we're after here)
If you like that, give unbound and ntpMerlin a go too.
There are reasons all of these are popular. I strongly suspect you (and your network users) will enjoy these additions.

 

[Morris]

Are there any issues on your network? So far you have demonstrated that the tests may not be reliable.

Morris

 

[Eric_on_Fire]

Are there any issues on your network? So far you have demonstrated that the tests may not be reliable.

Morris

Not any longer. I was having issues with dropped clients and packets, but all is seems good now that I've manually set it back up.

 

[Xrsenal]

(I dropped ConnMon in favour of SpdMerlin a while back - I didn't see the point in duplicating statistics/charts. also - zoom out (7d/30d) and switch to logarithmic scale for some perspective - it's not terribly bad)
You've just made the argument for yourself to use QoS - you have an asymmetric connection, where your download capability exceeds the upload significantly: the packets looking to "get out" have to look for spots to step into the traffic. by imposing a "speed limit" and "spacing rule" (2 car lengths kind of thing) you're going to see the consistency you're looking for, or much closer to it. (And yeah, MTU factors in to the equation not insignificantly.)

Could you post which hardware and firmware versions you're using please? Also - have you set up a swap on your entware drive? that'll help too, along with ntpMerlin

What if my speed is 900 upload and 500 download. Do I rate limit so it’s 500/500?

 

[L&LD]

900 upload?

You would rate limit each until the connection improved from your end. There is no 'correct' absolute amount to suggest here.

 

[Xrsenal]

900 upload?

You would rate limit each until the connection improved from your end. There is no 'correct' absolute amount to suggest here.

Yes my fios has 900 upload speed and 500 via dsl reports (Bc if multiple server)

but ookla is like gig both ways

 

[Xrsenal]

Your ISP paid-for speeds are what you state when someone asks for your upload/download speeds.

Don't base your responses on websites that obviously have underpowered servers.

Stop relying on misleading/bogus online tests and benchmarks. Networking (at least wired) isn't black magic. It is more scientific than that.

Appreciated. Can you take a look at my moca post I tagged u in ?

 

[L&LD]

I don't have any good input for you with Moca, (sorry). It is easier to run Ethernet than tangle with Moca setups for me.

 

[L&LD]

And for how non-linear setting up QoS is, have a look at this post below.

Wi-Fi Setup 25/2.5 d/u Report