When discussing next-generation architectures with our Service Provider customers, I am often asked  how they can have silicon diversity for roles in their network, but maintain common operational model, carrier-class features and performance.

The IOS XR offers just that. It runs on Cisco and merchant silicon across a rich variety of form factors. In a previous blog, I discussed the ASR 9000, today I’d like to highlight how IOS XR works some of its magic on merchant silicon.

Before getting into the details, let me share that NCS 5000 and NCS 5500 are actually not the first systems to run IOS XR on merchant processor ASICs. Other routers in our portfolio also use discrete merchant components, so the team has long experience developing software capabilities beyond what is commonly available in standard libraries.

As an example of IOS XR innovation and fine tuning that extracts maximum performance from forwarding silicon for real-world use cases, my team has been working hard on fast convergence since we first shipped the NCS 5500 series, and now we are proud to share the results – a 10-fold improvement in fast convergence. This means our customers have Carrier-Class fast convergence characteristics on the Cisco NCS 5500, thanks to the performance of and optimizations in IOS XR.

Need for Fast Convergence in Merchant Silicon

Many merchant network processors were initially designed for data centers. So, specific ASIC characteristics such as table sizes and forwarding capabilities were focused on switch use cases. For example, Forwarding Information Base (FIB) sizes were often limited to 32-128K entries. At that scale, how fast you program the ASICs forwarding tables isn’t critical.

With NCS 5500, however, we support a broad set of use cases and multiple roles in the Service Provider network, so we had to address some of these silicon design trade-offs. Coming back to the FIB size, we extended in two ways – adding an external TCAM, which enables up to 10M routes on some cards and optimizing label storage on-chip.

Now, with the increase in hardware FIB scale, there was a need for faster programming of the forwarding tables, which is the key platform-specific component of fast convergence.

In response, the team developed a comprehensive set of innovations in our forwarding code, specifically how routes and labels are organized in software and downloaded by IOS XR into the silicon.

Thanks to this work, we were able to deliver a 10-fold improvement in Fast Convergence times in real world deployment.

This has already proved beneficial to our customers. One of our major SP customers made the decision to build their next generation mobile core network with Cisco Segment Routing technology on the NCS 5500 series. Because they wanted their new infrastructure to support upcoming 5G use cases, they needed sub-second convergence with SR and LDP interworking. This was a great opportunity to apply our innovations and demonstrate that we could deliver the type of sub-second convergence the customer was used to, but on merchant silicon.

Thanks to the power of the IOS XR, we have amazing convergence performance on the merchant silicon based NCS 5500 router portfolio. While merchant silicon presents new and different design trade-offs, this is another example of how IOS XR delivers the best system and software innovation enabling us to get the best out of our customers’ choice of silicon.


Sumeet Arora

SVP Engineering

Core Software Group - US