We just upgraded our internet to 940/940 Mbps and installed a CRS326-24S+2Q+RM with it.
The whole office is wired in cat6a, and computers has 10Gbps NICs. Router has copper SFP+ 10Gb 300m transceivers.
The depot is on a QSFP 40Gbps.
The router reports the port to internet is a duplex 1Gbps. Which is normal, since our ISP modem port is 1Gbps.
The router reports others ports - to computers - to be 10Gbps and the depot to be 40 Gbps.
Doing iperf between a computer and our data depot results in 3.2/3.2 Gbps.
Doing any speed test with the internet results in 320/320 Mbps.
Basic error: CRS326 is not a router, it’s a switch. Yes, if it’s running ROS, it can route, but it does so using its slow CPU, so routing is nowhere near wirespeed.
Unless you’re running ROS v7 (not recommended in any kind of critical environment as v7 is still functionally incomplete and pretty buggy) which can offload certain routing tasks to HW on CRS3xx in which case routing can be wirespeed.
I’ve no idea. Perhaps MT marketing department has an answer to that? It does come with word combination router switch though … which should ring a bell, but it sure is hard to find out what it’s supposed to mean. At least for somebody new to Mikrotik universe.
No, IMO MT doesn’t have a good stable router with QSFP ports right now. They have only one router (CCR2216-1G-12XS-2XQ), but it’s a very new product and still suffers from teething problems. If you manage to get hold of it at all. It’s promissing though.
The CRS326 line is listed on the MT product page under switches. That’s this product’s primary purpose. If it was primarily a router, it would be listed under routers.
The CRS line does both, but more switching than routing. They’re excellent for cases where the bulk of the traffic is switched, but you need some L3/L2 intelligence in there for a subset of the traffic. I use mine primarily as a WireGuard gateway and a multicast router for the LAN. Non-WG and non-IGMP traffic is switched down at the hardware level, leaving the CPU peaking at about 5%.
The product you picked has a single-core 650 MHz CPU. You simply can’t push 40 Gbit/sec with that. Even something like the CRS5009 isn’t likely to fill a 10 Gbit/sec pipe with its quad-core 1.4 GHz CPU. That’s why the higher-end routers in MT’s line have even more cores than this. When most of the traffic goes through the CPU as in proper “routing,” you need a lot of CPU power to push many gigabits per second.
Is there a good router with a QSFP?
In your situation, I’d delegate Internet routing to the ISP modem, then run the CRS326 in switching mode, adding intelligence to the network beyond what the presumably-crappy ISP hardware offers.
If you need LAN-side routing at tens of gigabits, a big old honkin CHR might be your best option.
These units are not really different conceptually to a Layer-3 switch from Cisco e.g Catalyst 3560G( one sitting on my desk atm ! ).
I have 1xCRS326-24S+2Q+RM in my network, but its primary role is wire-speed switching, and the only offloaded cpu task I hand to it is the management vlan network.
Probably better to put something grunty up front like a CCRxxx or RB5xxx ( Cisco ISR ! lol ) “Router on a stick” type of config, and pass back traffic back into that CRS switch via VLAN’s for distribution.
It’s a video game production studio. I need a lot of router options, so the crappy ISP box is currently in bridge mode.
As for switching, we are using a Cisco N9K-C93128TX. Crazy how used one are so cheap on ebay. (Paid 550$ for our)
I guess I was thinking that handling a 1Gbps internet connection was the most basic thing.
Was also hoping to shield myself when we will move the internet connection on higher speed.
Correct me if I’m wrong, but I think you only need 1Gbps on the WAN side. You could use RB4011iGS+RM for that. If you look at the performance test results here https://mikrotik.com/product/rb4011igs_rm#fndtn-testresults then you will see that it can almost always route more than 1Gbps. The RB4011 can use routeros v6, that is stable. I would connect that router to the switch with SFP+, and your ISP device to the switch with QSFP (if that is what you have). Then configure the router as “router on a stick”. It will be able to route traffic to WAN at 1Gbps. It won’t be able to route much more than that, but if I understood your setup then most of your inter-LAN traffic will only need switching, not routing. (I might be wrong, if you plan to use multiple VLANs and heavy inter-vlan traffic on your LAN, but that is not likely.)
That works if the main “routing” needing to be done is to/from the Internet, and if the ISP modem has no useful amount of intelligent routing of its own. It lets the OP move the CRS326 into a network core role, where it’s much better used, distributing traffic on the router-on-a-stick fashion as you say.
Another nice feature of this plan is that it gives the OP another 9 GigE ports to play with, since the uplink to the core would be by fiber. Those ten GigE ports then get consolidated to the 10G link back to the core by the CPU, which means you get another opportunity for intelligent RouterOS filtering, since hardware offloading doesn’t get in the way.
There are people who have been using MikroTik for over a decade, who have built up complex configurations that use every little corner-case and expect it to behave in 2022 precisely the same way it did in 2012. Only with less bugs. And more features. And on the same hardware as in 2012. Without getting bigger.
And then there are people who have been doing networking for the same amount of time and who arrive at MikroTik only here in 2022 and say, “Hey, this v7 thing is pretty awesome as it is.”
Who’s right?
I urge a pragmatic approach: does your use case work on v7? Yes? Okay, v7 is awesome. For you.
What do you care if v7 isn’t awesome for everyone else on the planet?
Everyone has an opinion, and every use case it’s different. If you can, give it a try. If it fits you, ok. If it doesn’t, well, you’ll be as bad as now.
I think that your configuration is simple and probably will work well. But we won’t know until you test it. Remember to use hw offload for L3
