According to LightCounting/Téral Research, the global 5G Open Virtualized RAN (vRAN) market is set to triple in size by 2027. Mobile Network Operators (MNOs) are now deciding between two architectural approaches to the accelerator aspect of mobile networking: inline and lookaside. Today’s decisions will be the building blocks for tomorrow’s mobile networks, so the choice between the two has important implications for years to come. In this blog, we’ll dive into the inline vs. lookaside question and explore the advantages and disadvantages of each architectural approach to help operators understand which is best for different situations.
The power of vRAN disaggregation
Traditional RAN relies on purpose-built appliance hardware for baseband processing. By contrast, vRAN separates hardware from software, which offers MNOs unprecedented flexibility. They can now utilize Commercial Off-The-Shelf (COTS) hardware from various manufacturers. This is a seismic shift that not only translates to better pricing and faster innovation, but also enhances supply chain resiliency. Moreover, vRAN’s hardware-software separation unlocks the benefits of cloud-native solutions, such as Continuous Improvement and Continuous Development (CI/CD) and containerization. It also enables pooling of hardware for scalability, reduced capex, optimized power consumption, and increased redundancy.
The role of accelerator cards in Open vRAN
As Open RAN networks evolve, high-performance computing power that is also energy-efficient and economical becomes non-negotiable. Layer 1 processing at the Distributed Unit (DU) Layer 1 plays a pivotal role in orchestrating complex algorithms including Forward Error Correction (FEC), channel estimation, modulation, and layer mapping. Accelerator cards bridge the gap between these intricate tasks and the limitations of conventional General-Purpose Processors (GPPs) .
Understanding inline vs. lookaside accelerators
There are two main types of accelerator architectures: inline and lookaside. Each architecture employs accelerator cards that serve different functions.
- Inline Layer 1 accelerators: In contrast, inline architectures intercept Layer 1 processing for user plane data before it reaches the CPU. Inline Layer 1 accelerator cards can efficiently handle almost all Layer 1 processing tasks, freeing up CPU resources for Layer 2 and Layer 3 processing.
- Lookaside FEC accelerators: In the lookaside architecture, the CPU acts as the master controller for Layer 1 processing, with an accelerator card, often a separate PCIe card, handling select functions like FEC. However, many real-time computations continue to be processed by the CPU, including Layer 2 and Layer 3 processing.
Caption: Comparing inline and lookaside architectures
The choice between inline and lookaside has a significant impact on vRAN performance. Here’s a comparison:
- CPU resource allocation: With inline Layer 1 accelerators, user plane data remains exclusively on the accelerator card, reducing CPU resource consumption. In contrast, lookaside architectures route all user plane data through the CPU.
- Processing delay: Since lookaside architectures require many Layer 1 processing functions to be performed on the CPU, they are more likely to exhaust the more rapidly accessed, limited, primary or secondary cache, and they more frequently need to use tertiary cache that takes more time to access. Thus, lookaside architectures often result in higher bus utilization, which can lead to processing delays. While the delays associated with lookaside architectures are sufficiently low to satisfy O-RAN fronthaul latency budgets, this may limit new more latency-sensitive functionality like Fujitsu’s Extended Fronthaul capability that enables MNOs to extend the range of RUs up to 50 km from the vDU.
- Capacity: The additional delay of lookaside architectures can also limit capacity in multiple ways. Lower Layer 1 processing capacity limits the number of cells a vDU can support. In contrast, not only can inline accelerators support higher Layer 1 processing capacity, but by taking over the bulk of Layer 1 processing, their general-purpose (and more expensive) CPUs are also freed up for more critical tasks like Layer 2 and Layer 3 processing. This resource optimization enables you to support more traffic, more User Equipment (UE), and more connections—without the need for more costly and power consuming servers.
- Energy efficiency: Inline architectures tend to be more energy-efficient, especially in scenarios with many cells. However, in small sites with few cells, low traffic, and limited spectrum, lookaside architectures may be more energy efficient.
Unlocking the potential of inline Layer 1 accelerators
Inline Layer 1 accelerators offer several other advantages:
- Flexibility: With inline accelerators, Layer 1 capacity can scale independently from Layer 2 and Layer 3, allowing MNOs to select server hardware optimized for different deployment scenarios.
- Cost optimization: With the flexibility to scale the layer 1 processing separately from Layer 2 and 3 processing, MNOs can select CPUs so that Layer 2 and 3 processing capacity is balanced with Layer 1. Balancing the capacity minimizes excess computing capability so that MNOs only pay for what they need.
- Large vendor ecosystem: Embracing solutions that support standardized Functional Application Platform Interfaces (FAPI) and allowing for multiple vendors promotes flexibility, innovation, price competition, and supply chain resiliency. The ability to mix and match components from different vendors aligns with the core principles of virtualization and avoids vendor lock-in. While there are several vendors continuously improving to compete for the inline accelerator market today, there is currently only one vendor offering a lookaside FEC accelerator.
Additional advantages of lookaside accelerators
While inline Layer 1 accelerators offer compelling advantages, lookaside architectures may find their place in specific scenarios:
- Simplicity: Lookaside solutions, typically offered by a single vendor, provide an all-in-one, easy-to-manage approach suitable for small, low-traffic sites.
- Inventory Management: A single-vendor “all-in-one” approach reduces the complexity of managing different hardware components and development cycles.
The bottom line: choose the best accelerator for your network
In the ever-evolving landscape of Open vRAN vDUs, the choice between inline and lookaside accelerators is an important decision for MNOs. Inline Layer 1 accelerators offer a lower total cost of ownership (TCO) by optimizing resource usage, improving performance, and increasing capacity and flexibility. This type of accelerator is particularly well-suited for scenarios with medium to high traffic.
While lookaside FEC accelerators may find their niche in low-traffic sites, the broader ecosystem of inline accelerators, supporting multiple vendors and increased competition, provides MNOs the tools they need at the right price to stay agile and competitive in the ever- changing world of 5G.
Fujitsu recognizes that one size does not fit all, so we support accelerator cards—as well as servers and CPUs— from multiple vendors. We’ll continue to grow our ecosystem of partners, prioritizing support for new hardware based on market demand.
Ultimately, the path chosen will influence not only immediate performance but also an MNO’s ability to adapt, innovate, and thrive in the rapidly evolving 5G landscape. Inline versus lookaside is a choice that will shape the future of mobile networks.