I’m using MikroTik routers and switches in my WISP network (Mimosa radios). Overall, they work well. But I think I may be getting some bottlenecks in the network.
I have a backhaul network, using switches, and then each site as a ‘site router’, that connects to the switch that is at that site. The access points connect to the site router on the LAN side.
In some cases, the backhaul network doesn’t ‘hop’ to another site from a site, so their I have no switch and the backhaul network connects to the WAN port of the site router.
I’ve been using CRS326-24G-2S+RM Cloud Router Switches, running RouterOS (6.49.6 firmware), with no settings in it such that it should be offloading the data to hardware.
But now I’m wondering… Will the same hardware (CRS326-24G-2S+) handle traffic any better (more traffic, faster, etc) if it’s running the SwitchOS firmware as compared to the RouterOS firmware?
Switches move traffic around at wire speed because of the switch chip (ASIC), provided they are configured correctly (one bridge, all ports in the bridge, VLAN filtering enabled and VLANs assigned to their respective ports). The CPU should show hardly any usage except when you browse into it, run tests, and so on. There will be no difference in networking speed between RouterOS and SwitchOS (except SwitchOS will keep you from using most RouterOS features that would hit the CPU).
The faster CPUs in some of the newer switches just means they’ll be slightly better at handling ancillary tasks like management or routing protocols, which is great if you use Layer 3 Hardware Offload and use them as routers and not just switches.
If you’re just switching at Layer 2, the CPU speed won’t make a difference because it’s not involved.
I see what you guys mean. Whether it’s running RouterOS or SwitchOS, it’s using the same hardware, assuming that the RouterOS is configured so that Hardware Offloading is functional.
The only way to something faster is to change hardware.
That was my point. You haven’t explained why you think the bottleneck is the switch.
What I am saying is that usually a switch ASIC is not the bottleneck. Most switch ASICs have the capacity to handle full speed connections between mutually exclusive port pairs, e.g. if you had an 8 port switch with ports numbered 1-8, if you paired each odd numbered port with the next higher numbered port, each pair should be nearly wire speed, full duplex. Look at the test results MikroTik published. This was already said by @sirbryan in post #4
Switch bottlenecks are usually a trunk link where things get funneled into a saturated link. Even more common is not even related to the network, e.g. an application server.
If you are at the point of saturation that a switch makes a measurable difference, then you need a switch with faster links/backplane.
If all that is being done is L2 switching, and you have ROS configured correctly so that the switching of offloaded from the CPU, I don’t think that the switch is the problem.