AiMesh Adventures including Stock v Merlin on node

[Morris]

I see. Thanks.

I already have 4 asus routers in the house so not planning to buy any more

Looks like I have to be happy with what I have or run some cables to link the nodes

I wish I had better news. I just finished my second cable run though I can't hook up right now as my family is using the network and don't want to cause an outage if I goof or Asus code strikes again. The router sometimes crashes when the network converges though I think that's only on WiFi. I don't want to find out right now.

 

[jata]

Last update (for now)...

Using iperf3 to test/monitor I have found that i could improve speed and stability by moving the node's location in the study/bedroom. Best spot was actually close to my mac mini that I use a lot and it is also my media server - this was the client where I really needed optimum network performance.

Even better news is that now the node is next to my mac, I can connect it to the node wired - the node is still wifi backhaul but the improvement is significant.

• test1: macmini wifi 5Ghz to node at 3m away = 20MB/s (160Mb/s)

changed location of node in the room

• test2: macmini wifi 5Ghz to node at 0.5m away = 30MB/s (240Mb/s)
• test3: macmini ethernet = 50MB/s (400Mb/s)

Whilst 400Mb/s is not amazing - it is a 250% increase

So my advice to everyone with AiMesh is really simple (and probably obvious too)

1. ethernet backhaul where possible
2. use iperf to find best location for node within a room - connect iperf client to LAN port on node for testing
3. where possible wire clients to the node - you might be better to lose some wifi backhaul speed to be able to wire clients (e.g. servers/workstations) to a node

 

[John_]

@jata Do you have cable tv coax wiring in the rooms with the router and the nodes?
If so, you might consider getting MoCA adapters and run the Ethernet backhaul that way.

 

[jata]

@jata Do you have cable tv coax wiring in the rooms with the router and the nodes?
If so, you might consider getting MoCA adapters and run the Ethernet backhaul that way.

Thanks for the suggestion. There are some old coax points around but I have never used them and none of them are at the right location.

I'm quite happy now with 50MB/s from my media server to my main router and NAS across the wifi backhaul - that's fine for my home use.

This investigation has highlighted that I might have another issue/constraint. At the moment, media data runs from NAS (main router) --> media server (node) --> client (many places) all across the wifi backhaul on the node. It would probably be better if I connected my NAS to the same node as my media server / mac mini. But this is getting too messy (and the wife does not like to see the 'messy' tech).

Anyhow, it's 90% streaming services across the WAN/internet (100/40) these days for our family so not really an issue.

 

[JGrana]

Great thread, lots of good information for us with AiMesh systems.

As a side, I did notice that on the HND class routers I have (AX88U, AX58/AX3000 and AX86U) running the latest Merlin code and one of the AX3000 running stock - that the iperf3 binary is in /usr/bin. As a result, you don’t need to load Entware and install the iperf3 package from the repository.

You can get it running in server mode in the background by ssh’ing into the router and doing:

iperf3 -s -D

You could enable it on all the nodes and test node to node bandwidth.

Unfortunatly it doesn’t survive a reboot… There are some nvram variables that appear to allow this (i.e. iperf3_svr_enable and iperf3_cli_enable) but setting either or both to “1” doesn’t result in the server starting at boot time.

I checked on an older AC68U running the latest code - no iperf3 included.
Also, the iperf3 binary from Asus is a few minor revs behind the Entware version. Not sure it matter, I have done some testing (node to node) with both and the results are pretty much the same.

 

[Morris]

Last update (for now)...

Using iperf3 to test/monitor I have found that i could improve speed and stability by moving the node's location in the study/bedroom. Best spot was actually close to my mac mini that I use a lot and it is also my media server - this was the client where I really needed optimum network performance.

Even better news is that now the node is next to my mac, I can connect it to the node wired - the node is still wifi backhaul but the improvement is significant.

• test1: macmini wifi 5Ghz to node at 3m away = 20MB/s (160Mb/s)

changed location of node in the room

• test2: macmini wifi 5Ghz to node at 0.5m away = 30MB/s (240Mb/s)
• test3: macmini ethernet = 50MB/s (400Mb/s)

Whilst 400Mb/s is not amazing - it is a 250% increase

So my advice to everyone with AiMesh is really simple (and probably obvious too)

1. ethernet backhaul where possible
2. use iperf to find best location for node within a room - connect iperf client to LAN port on node for testing
3. where possible wire clients to the node - you might be better to lose some wifi backhaul speed to be able to wire clients (e.g. servers/workstations) to a node

Excellent information. I've observed the same. As people design, they should remember wireless is for convivence and wired for performance. If you can avoid a wireless connection or hop there will be better throughput and reliability.

 

[Morris]

Thanks for the suggestion. There are some old coax points around but I have never used them and none of them are at the right location.

I'm quite happy now with 50MB/s from my media server to my main router and NAS across the wifi backhaul - that's fine for my home use.

This investigation has highlighted that I might have another issue/constraint. At the moment, media data runs from NAS (main router) --> media server (node) --> client (many places) all across the wifi backhaul on the node. It would probably be better if I connected my NAS to the same node as my media server / mac mini. But this is getting too messy (and the wife does not like to see the 'messy' tech).

Anyhow, it's 90% streaming services across the WAN/internet (100/40) these days for our family so not really an issue.

Possibly you can move your media server near the node where it will be accessed

 

[heysoundude]

Excellent information. I've observed the same. As people design, they should remember wireless is for convenience and wired for performance. If you can avoid a wireless connection or hop there will be better throughput and reliability.

This Is The Way.
Until you've tried it for yourself to see the difference, you don't know...and knowing is half the battle.

 

[Morris]

I completed migrating to an ethernet backhaul, yet the adventure continued. First I had a bad cable so I pulled a replacement. Then I connected it to my RT-AX86U node and while the link light came up, the ethernet backhaul did not. I attached my laptop to the cable and had connectivity. There is a dumb switch in the bath between the node and the router so I went to the uplink to the dumb switch with a node and the ethernet backhaul came up. The switch is an old Asus 5 port dumb switch. I replaced that switch with a different model and at last an ethernet backhaul with improved throughput and fast network converging. I waited a few hours and then went back to auto channels and enabled 5-Ghz/160 with DFS. Aircraft have flown over and the channel even changed yet this time no router crash!

Wired like this the setup reminds me of a cooperate WiFi setup with controllers that distribute settings. There is one big difference: all the APs are on the same channel causing cross channel interference. If Asus automatically assigned channels or let us do it we would have a faster network.

 

[jata]

Ran some tests streaming HD films from media server (plex) to my apple TV over the 2-way trip across the node and even with the film being transcoded rather than direct play, the data bandwidth is not an issue.

Another thing I found is that iperf3 on a NAS connected (1Gb) to my router produces faster and more consistent results than running iperf directly on the router. This surprised me and I migght test this a bit more (i had a few wines last night when running the test.

So I'm generally happy with my setup as-is but would like to run some ethernet cables for backhaul at some stage...

Thanks all for the input and suggestions.

 

[tgl]

Another thing I found is that iperf3 on a NAS connected (1Gb) to my router produces faster and more consistent results than running iperf directly on the router.

Yeah, I have observed the same effect with my Ubiquiti EdgeRouter X. There is a built-in iperf3 server, but it seems to max out around ~860Mb/s when the router is sending, ~630Mb/s when receiving. This despite the fact that the router can pass full wire rate (~935Mb/s) between its ports. I suppose that the port-to-port case is aided by some packet-forwarding hardware that isn't used when the router's CPU is source or destination. Given that the CPUs in these things are on the wimpy side, it's not so surprising that the CPU can't keep up.

 

[Morris]

Since implementing AI Mesh, I've had the router crash on some settings changes. Today it was when I put an additional MAC address filter on the 2.4 Ghz band. This was because my laptop would not connect to 5-Ghz. Did not fix the issue which I determined was the laptop properly refusing to connect to 36/160 as half is DFS and there were aircraft in the area. Asus vitiated this. Removing the MAC filter also crashed the router. I then changed to 5-Ghz/80 no DFS and that changed without a crash and my laptop connected. I may stay at /80 as /160 dose not result in much additional throughput. I wonder if the /160 was confusing the nodes?

 

[jata]

Since implementing AI Mesh, I've had the router crash on some settings changes. Today it was when I put an additional MAC address filter on the 2.4 Ghz band. This was because my laptop would not connect to 5-Ghz. Did not fix the issue which I determined was the laptop properly refusing to connect to 36/160 as half is DFS and there were aircraft in the area. Asus vitiated this. Removing the MAC filter also crashed the router. I then changed to 5-Ghz/80 no DFS and that changed without a crash and my laptop connected. I may stay at /80 as /160 dose not result in much additional throughput. I wonder if the /160 was confusing the nodes?

hey Morris - I reverted to 5GHz/80 for the same reasons. It's a hassle having to deal with DFS and the performance gain of 160 was not really faster even though the nodes 'seemed' to be connected at a higher speed the actual throughput was the same or lower than 5Ghz/80.

My network is quite stable and consistent across the nodes so I'm not changing/tweaking anything

 

[Morris]

Since implementing AI Mesh making simple changes on the router resulted in crashes. For example adding or removing a Wireless MAC Address Filter usually crashed my router. Today I found a work around: In the AI Mesh Settings set Ethernet Backhaul Mode to ON. Of cause you need an ethernet backhaul

 

[jata]

Thought i'd update this thread as it's a good one and this story is 'on topic' (sort of)

Decided to go 'all in' and migrated main router (AX86S) to merlin 386.7 B1 from 386.5 and upgraded the nodes from stock to merlin at the same time... It's been a rocky couple of days at home with kids addicted to tiktok and wifey working from home

I did the nodes first and that all went well so I can confirm that you can update from stock to merlin on nodes through the GUI and all seemed fine. I SHOULD HAVE WAITED BEFORE PROCEEDING TO MAIN ROUTER but i got over confident and trigger happy...

Updated AX86S to 386.7 about one hour before the issue with updating this model to 386.7 was reported... Fortunately the update was ok but I did have the jffs issue that I fixed using a jffs backup.

But things were not perfect and had a few issues with my homepods not connecting intermittently so I did a full nuclear reset of my entire network main router and nodes. 3hrs+ later...

Same issue with homepods so I decided to go back to stock on nodes. This was a mission as I needed to use the restoration tool/process to get the nodes back to stock. Same issue with just one homepod on one node.

So another full nuclear reset of entire network with merlin 386.5 on main router and latest stock on nodes. and finally everything back to normal. I did factory reset all my homepods and moved the one playing up to a different location so it's possible this is a factor.

Anyway - main lesson(s) learnt from this:

• Don't rush a big update - do one node first and check everything working as expected. then next etc..
• Don't make a big change to entire network using beta software. I work in IT and I know this but I still did it!
• Don't run merlin on nodes unless you really need to (e.g. scripts/logging etc.) - I'm 100% convinced stock is less glitchy on nodes
• I can complete a full reset / rebuild of my network blind folded in around 3hrs

 

[Morris]

Thought i'd update this thread as it's a good one and this story is 'on topic' (sort of)

Decided to go 'all in' and migrated main router (AX86S) to merlin 386.7 B1 from 386.5 and upgraded the nodes from stock to merlin at the same time... It's been a rocky couple of days at home with kids addicted to tiktok and wifey working from home

I did the nodes first and that all went well so I can confirm that you can update from stock to merlin on nodes through the GUI and all seemed fine. I SHOULD HAVE WAITED BEFORE PROCEEDING TO MAIN ROUTER but i got over confident and trigger happy...

Updated AX86S to 386.7 about one hour before the issue with updating this model to 386.7 was reported... Fortunately the update was ok but I did have the jffs issue that I fixed using a jffs backup.

But things were not perfect and had a few issues with my homepods not connecting intermittently so I did a full nuclear reset of my entire network main router and nodes. 3hrs+ later...

Same issue with homepods so I decided to go back to stock on nodes. This was a mission as I needed to use the restoration tool/process to get the nodes back to stock. Same issue with just one homepod on one node.

So another full nuclear reset of entire network with merlin 386.5 on main router and latest stock on nodes. and finally everything back to normal. I did factory reset all my homepods and moved the one playing up to a different location so it's possible this is a factor.

Anyway - main lesson(s) learnt from this:

• Don't rush a big update - do one node first and check everything working as expected. then next etc..
• Don't make a big change to entire network using beta software. I work in IT and I know this but I still did it!
• Don't run merlin on nodes unless you really need to (e.g. scripts/logging etc.) - I'm 100% convinced stock is less glitchy on nodes
• I can complete a full reset / rebuild of my network blind folded in around 3hrs

Good advice!

I feel the easiest way to go from Merlin to stock on a node is to do a factory reset. Connect the node's WAN port to a port on the router. Connect a laptop to a port on the node and follow the startup wizard accepting all defaults. Once it comes up go to upload firmware and it will take it.

Morris

 

[jata]

Good advice!

I feel the easiest way to go from Merlin to stock on a node is to do a factory reset. Connect the node's WAN port to a port on the router. Connect a laptop to a port on the node and follow the startup wizard accepting all defaults. Once it comes up go to upload firmware and it will take it.

Morris

yep - that's what I would normally do (and it worked on my AC68U) but the for 2x AX58Us need the restoration tool to get them back to stock from merlin 386.5 (known issue).

Good news is that 48 hours of everything working great. I just have an annoying issue with clients disconnecting from my AC68U node in the garage. I have removed it from my setup as clients seem OK connecting to a AX node further away.