I was looking for a 12 or 16 port SFP gigabit managed switch for securing my 10 POP locations by a protected ring fiber network.
I came across http://routerboard.com/CRS212-1G-10S-1SplusIN
and the price seems amazing, but before I put the money on this, I have to be sure that it can really handle what I need.
My network is CO and then 10 tower locations. All the towers get their own dedicated core from the CO and it will eventually reach back to CO, so it becomes a protected ring.
Now, each tower needs to have a fiber switch to input its core and output it go ahead.
And we might have local customers connected through that fiber switch as well. So, basically each tower will pass traffic like 5Gbps or so and over 12000 customers.
So, can this switch really handle it ?
Or will the CPU burst out ?
Im taking the max assumption as 5Gbps and 12000 customers.
Switching will be wirespeed, anything that gets passed to CPU on this switch will not be [pay attention to “Performance test results”].
I am having a hard time picturing your topology, however. Where in your network would the switch(es) go? At each tower + the CO? The confusion is because you say 5Gbps and a ring topology. My understanding of a ring would be one in and one out, but these switches only have a single 10G interface.
5gbps ? I wouldn’t not think about CRS in such network. Use something:
better manageable. See the hell of options with strange names and meanings under Switch menu. Near no one option uses a name which is widely used in networking world. Simple things like VLANs are rather complicated here and takes too much time to configure them.
better capable of such load. CRS Switch cfg maybe wirespeed but I think you need something with more buffers and more reliable stats about the data (SNM counters for errors etc)
because it is an entry level switch. I have no experiences with them. I would go with 2960-x with LAN Base licence/feature set (Lan Lite is a shit). I don’t know how many SFP ports you need (and whether you need 10gbps SFP+). If you can live with 4xSFP 2960x-24TS should be OK.
Honestly, I would never use Mikrotik as a core switching platform. Winbox is nice, and RouterOS has a million wonderful features, but Mikrotik’s ability as a pure switch is very very bad.
I also dislike ring topology for ethernet, unless you’re using something like REP to do fast failover. (R)STP sucks at dealing with rings. Topology changes can be nightmarish.
I use Microtik as my core router and NAS router only.
And for switch, I currently use UBNT EdgeSwitch, since our current customers are all wireless customers.
But since we are building a fiber network, I was actually looking to find a better SFP port switch, which I could feed from Microtik 10G SFP port and then all the fiber gets into the SFP port of the switch and job done.
And I talked to many about the ring and everyone suggested to go for RSTP way.
But you said, its bad and would go with REP.
Now, I have no idea about REP.
REP is a Cisco proprietary sub-millisecond failover protocol that you use instead of spanning tree.
Brocade’s got a similar thing called RRP (I think).
RSTP is capable of doing loop prevention in a ring, but if your ring has very many nodes, it can take a while for a failure to cascade around the ring and for the new path to converge properly, and restores take a bit as well. I once worked for a company where the HQ was on the most distant node on a ring from the main network pop. It was common (happened at least once a week) for the connectivity to drop for 15 to 30 seconds while something flapped. I started calling them ‘netquakes.’
(the ring was not my ring, and I was not in any position to change anything about it - we had to just endure the netquakes)
Anyway, I just think that rings aren’t a good topology unless they’re running a fast-failover protocol like SONET (of course I’m not advocating SONET - that’s old tech), or else fast failover isn’t an issue. OSPF can take a bit to converge when deployed in a ring, too…
You don’t have to use Cisco’s solution - WDM gbic/sfp modules exist for lots of vendors.
You can also purchase WDM add/drop muxes and termination splitters from many vendors as well.
The key idea is that you can run up to 8 different wavelengths to make each pair of strands act like 8 pairs of strands. You then connect your sites in a ring (physically) but logically, the topology is hub-and-spoke, so each site is only one hop from the head end.
I would do something like this because I hate failover by nodes that are connected in a ring, but not designed for rings. This solution gives the best of both worlds - the efficiency and relatively low cost of installing a ring (vs. a mesh or star topology), yet avoids the headaches of a logical ring topology.
Industrial networks use rings extensively. Schneider Electric uses RSTP and promises 50ms recovery with their gear (must be tuned). Cisco has REP, Rockwell does DLR. Hirschmann does Hiper Ring and MRP. It’s a crowded space for ring redundancy protocols…
I am not advocating them, only presenting information that some markets for Ethernet use these a lot. Datacenters and office networks would not be a typical use case.
I have 9 POPs, so 12 core will be done, 9 core will start from CO AND drop its own dedicated core at each POP and others will go ahead, and in the end, all 9 will reach back office the same switch.
