As far as I can see the FS726T has "some" Level-3 features but not all, as per the response to my static routing question.
We have one main 726 (A) I guess you would call it a hub, which is connected to other buildings, main office has another 726 (B) ,(50m away) other links are twisted pair with ADSL like boxes. to overcome the > 100m distance to building (C) 350m away.
If I pathping to anything (D {another host like a PC} on A from B and C I get 1ms and 16 ms respectively. This I suspect is because the link speeds are...
A to B is 100 Mbps
A to C is possibly 5 Mbps.
On a true Level-3 switch MAC address filtering would keep A to C traffic (and beyond to (D) completely separate and possibly prevent A-C-D traffic from slowing down the 726 and its other traffic.
I suspect that this A to C traffic does in fact slow all the traffic on the main 726 (A)
or..
The 726 as a somewhat old model does not simply have enough grunt.
Slowing down would be a consequence of the Ethernet CSMA/CD basically slow links cause lots and lots of re-tries, even resulting in the whole Ethernet network coming to a halt.
Any confirmation or not ! of my theory appreciated.
Nothing - at least nothing on each switch alone (full end to end 100 Mb/s there), nothing below 100 Mb/s for the links between the switches, and just the one link that is 5 Mb/s - would limit more than what is required These are switches - neither hubs nor fat yellow Ethernet cables - handling the traffic based on L2 information, which is obviously all MAC based. No L3 functionality would be involved in such a set-up.
Scratch any ideas about Ethernet CSMA/CD from 1980 - each of these collision domains is limited just to each link, say between two switches, say between a computer and the switch. Anything else is pure switching.
As you seem to have mutiple IPv4 subnetworks in the set-up (each switch has it's own based on what I have read earlier on your posts). We don't know anything on how these IPv4 subnets are connected on L3 - probably all routed traffic (outside of each subnet) will run over some router, probably your Internet router connected to just one of these switches. Well possible, this is where our performance (of what is possible in a legacy Fast Ethernet network) is going lost.
The L3 routing on the switch does just allow using switching technology to route between locally connected subnetworks only.
A MAC filter would be just to restrict some access within the network - not what you are looking for here.
A clarification on the multiple subnets (on one specific site) As the FS726T does not support routing (only multicast) this is connected with an Ethernet cable but disabled a it was tested and didn't (as expected) work.
-----------
More info which would have made the original question too detailed.
Very useful to know that each connection, in-out i.e. through the switch is its own collision domain. A possible wrinkle on this is that say on the slow link at the far end there are three PCs and when they get to the hub switch (A) all the traffic is going to one connection *say an Internet connection) then on both the link to the hub, and on the link from the hub to the Internet connection there is contention and so the CSMA/CD does apply.
If there were a single host at each end to and from the switch fine, but multiple connections at both ends means CSMA/CD is still a significant problem.
Add more in that both the link to the hub (5 Mbps) and the Internet link (12 Mbps) the connection out of the hub (A) are slow. I'm sure in this case CSMA/CD still applies.
Add to this the office 50m away is on a switch to switch connection (single 100 Mbit Ethernet) and the office has multiple user PC all connecting to the Internet too, this makes the link from A to the Internet even more busy and subject to the limits of CSMA/CD i.e. lots of collisions and associated waits and delays.
Each connection to the switch being a single and isolated collision
domain would be fine if each port on a single switch had one host but as soon as any connection has a single host that has multiple connection i.e. from different ports then the link to say Internet or file server or NAS is still subject to collisions = delays.
Checking connections through an internet router I also see twenty connections when you simply open a browser, making contention even worse.
A clarification on the multiple subnets (on one specific site) As the FS726T does not support routing (only multicast) this is connected with an Ethernet cable but disabled a it was tested and didn't (as expected) work.
Very useful to know that each connection, in-out i.e. through the switch is its own collision domain. A possible wrinkle on this is that say on the slow link at the far end there are three PCs and when they get to the hub switch (A) all the traffic is going to one connection *say an Internet connection) then on both the link to the hub, and on the link from the hub to the Internet connection there is contention and so the CSMA/CD does apply.
If there were a single host at each end to and from the switch fine, but multiple connections at both ends means CSMA/CD is still a significant problem.
Add more in that both the link to the hub (5 Mbps) and the Internet link (12 Mbps) the connection out of the hub (A) are slow. I'm sure in this case CSMA/CD still applies.
Add to this the office 50m away is on a switch to switch connection (single 100 Mbit Ethernet) and the office has multiple user PC all connecting to the Internet too, this makes the link from A to the Internet even more busy and subject to the limits of CSMA/CD i.e. lots of collisions and associated waits and delays.
Each connection to the switch being a single and isolated collision
domain would be fine if each port on a single switch had one host but as soon as any connection has a single host that has multiple connection i.e. from different ports then the link to say Internet or file server or NAS is still subject to collisions = delays.
Checking connections through an internet router I also see twenty connections when you simply open a browser, making contention even worse.
Dear Burt,
Afraid, I must disagree again.
Please avoid the hub term - this would be a classic Ethernet hub behaving like a fat yellow Ethernet cable with AUIs resp. a thinwire cable with T-pieces and multiple devices. Here the collision problem might apply - but not on a switch: Simply said on a switch with a full duplex link there is always only one sender in one direction, and this is on a dedicated links, a single cable pair in the case of Fast Ethernet for each direction.
Some reading (from a low cost device competitor - I hope Netgear is fine with that): What is the Difference Between An Ethernet Hub and a Switch? Interestingly it's dated from 2018 - but if you consider that you have these Netgeaar Fast Ethernet switches for a very long time you had been on leading edge tech already. Netgear stopped making hubs very early, in favour of switches, while low cost vendors continued offering hubs.
Needless to say:
Already the 5 Mb/s DSL link (bridged or routed - and is this some kind of symmetric like SDSL with 5 Mb/s in each direction) is very weak. What was probably fine when this was designed and implemented a decade or two ago for simple process communication, manufacturing stuff, probably computer terminal based system.
To me it's still a black hole on how your switches/sites/IP subnetworks are communicating.
I understand that you tried one of the switchss to act like a core or root switch where you wanted to link up the other switch or switches with a direct 100 Mb/s Ethernet link each. Again from my understanding that you have dedicated IP subnetworks on each switch/location, simply linking the same "network" together into one won't make the communication work magically. One one hand you interconnect multiple subnets whee probably each should have it's own DHCP server and subnet, on the other hand the some routing must be in place somewhere. Of course you could link up these L2 networks together, but to keep the different subnets isolated, VLANs must be configured, so the connection beteen the switches can act like a trunk carrying multiple (IP subnet) networks.
And of course, when you check the connections on the router, with an active user like me (having dozens of browser tabs, multiple brosers, and other applications open) there won't be just some ten concurrent connections open, there are easily hundreds. This has an impact on the available Internet bandwidth 12 Mb/s Internet [downstream, upstream is much less, typically 5..15% only] reads like a relict from the past. As long as your users are just using chat messaging and e-mail, this might be fine. Already the access of an average single modern Web page this ends in www (world wide waiting).
Certainly workable technically, your existing network is what I used to build in the late 1980ties or early 1990ties. Perfectly fit for purpose and requirements from back then. However, times have massively changed.
While I'm still convinced some short term tuning of the existing hardware by using Ethernet interconnections (with VLAN trunks), some clever routing for the subnetworks, could help to some better user experience.
Depending what your organisation or business does really need - hard to answer without any insight - thee is no other way than update your network. About the "bare minimum" I would suggest is using fiber pairs to interconnect the buildings/sites, this would allow Gigabit Ethernet or even affordable 10 Gigabit "backbone" connections, probably look towards at least Gigabit coverage for the end points (computers) especially if you intend to do some local storage for work and backup. Last but not least, if there are many active Internet users on this network, an update for the Internet connection is required.
