RB1000 vs L3 switch routing performance

RouterOS General
1

I wonder if anyone has any experience or knowledge about how a software based router like a Mikrotik RB1000 performs compared to a L3 switch or a “real router” when it comes to pure routing performance.

Benchmarking pps and Mbps through different devices with single or only a few simultaneous connections is easy and shows that a L3 switch like a HP Procurve 3400/3500 or a Cisco 4948 easily outperforms both a RB1000 and a “real router” like a Juniper M7i.

What I’m interested in though is how they compare when are used in more real-world environments as core routers in a small ISP network. I’m not talking about lots of Mbps then, perhaps 500-1000 Mbps of aggregated traffic, but with lots of different source/dest addresses, and perhaps some lower capacity links which gets congested and requires buffering for optimal performance.

A few years ago L3 switches suffered greatly in this scenario because of small host routing table sizes (fast-path caches) for ASIC routing, forcing them to start routing traffic in software when the hardware host routing table got filled up which then led to terrible performance.

Is this still a big problem with modern L3 switches as core routers?

How does a software router like the RB1000 compare to these. The RB1000 has significantly lower pps/Mbps ratings compared to modern L3 switches on the spec paper, but what about when you load it with traffic from say 200.000 different IP addresses. Which will perform best then?

Anyone done any benchmarks with lots of different addresses?

Any suggestion for opensource tools to simulate real world core router traffic patterns to benchmark this?

2

“real routers” :slight_smile: lol

Compare the price range!!!

They cannot be compared, they are completely different league, but If you will give me same amount money that you spend on brand new “real router”, I will be able to create a whole medium size network from RouterBoards (also CPE included)

3

Backplane of L3 switch is very fast, and it can handle a heavy load of traffic. But you cannot compare things like that, everything depends on your needs and budget. The one of the cheapest L3 switch is Cisco 3560, and it costs with 24 ports around 5000$ and 48 ports around 9000$, where the RB 1000 costs only 600$, and it is a completely different architecture. I have experience with Cisco Catalysts 6509, 4506 and 3560 L3 Switches (I have few of these up n’runnig), and I may say that the performance of these switches is amazing, but they have its limitations where you will have to use other hardware aswell, because of that I had to involve some Cisco ISR 3845 and 2851 and also few RB-1000.

Regsrds

Faton

4

My understand of l3 switches is that most of the affordable ones do not have the means to handle a full internet bgp routing table. They’d be fine for internal BGP if they even do BGP. They would probably not be calea compatible too, apart from port mirroring to a router, which is important for US providers.

The best of both worlds is a good router with tagged vlans over to a good managed switch to get the port count you need.