Advanced Switching

Switch

Advanced Switching was a PCI Express-based switched fabric specification developed in the early 2000s for communications, storage, embedded, and modular computing systems. It was not an Ethernet switch and not a software switching feature. It was an attempt to use PCI Express physical and data-link technology as the foundation for a richer system interconnect with packet-style fabric behavior.

The goal was ambitious: take the speed and silicon ecosystem of PCI Express, then add features that telecom and embedded-system vendors wanted in backplanes and modular systems. Those features included peer-to-peer communication, quality of service, multicast, high availability, fabric discovery, congestion management, and support for protocol-agnostic data movement between boards and endpoints.

What Advanced Switching Tried to Add

Switched fabric behavior: multiple endpoints could communicate through a fabric rather than only through a host-centric tree.
Peer-to-peer traffic: endpoints could communicate directly without forcing all traffic through a central processor.
Quality of service: traffic classes and scheduling were intended to support latency-sensitive communications workloads.
Multicast: one-to-many communication mattered for control planes, line cards, and distributed systems.
High availability: hot add/removal, redundant paths, failover, and fabric management features were aimed at carrier-grade equipment.
Protocol flexibility: the fabric was meant to carry different upper-layer protocols over a common switched interconnect.

Why It Was Interesting in 2003

Communications equipment was changing quickly in 2003. Vendors were building blade servers, AdvancedTCA shelves, storage systems, routers, media gateways, and packet-processing platforms that needed more bandwidth between boards. Proprietary backplanes were expensive and fragmented. Ethernet was strong, but it was not always the preferred internal system fabric. PCI Express was emerging as the next general-purpose serial I/O technology, and the Arapahoe Working Group tried to adapt it for a broader fabric role, in parallel with the system trends behind 10 Gigabit switching.

Intel's 2003 announcement described Advanced Switching as a communications-oriented complement to PCI Express: a switched interconnect and data fabric technology for joining components and system boards in low-to-midrange communications and embedded applications. EDN described the specification as supporting high-availability capabilities such as hot adding and removal of boards, redundant pathways, management failover, peer-to-peer unicast, and multicast communications among endpoints.

The 2003 Sandburst Announcement

In 2003, Sandburst Corporation joined the Arapahoe Working Group. The group was developing the Advanced Switching interface specification for switched data interconnection in storage, computing, and communications equipment. Advanced Switching provided advanced communication features on PCI Express technology.

Sandburst planned to support development of the specification. Its expertise in high-performance packet-switching architectures was expected to contribute to the Advanced Switching effort. Sandburst's role in storage and blade-server standards, including iSCSI and RDMA over IP, was also positioned as useful for developing Advanced Switching as a platform for modular storage, computing, and communication systems. Sandburst planned to support Advanced Switching-based equipment in future products.

"We look forward to sharing our quality of service (QoS), traffic management and packet switching expertise in the AS standardization effort," said Stephen Bailey, chief technologist of Sandburst Corporation. "The broad adoption of AS will reduce development costs and accelerate delivery of next generation modular systems."

Arapahoe Working Group members included Intel, Agere, IBM, Integrated Device Technology, Vitesse Semiconductor, AMD, Agilent Technologies, Huawei Technologies, Alcatel, Siemens, Sun Microsystems, EMC, Fujitsu, Hitachi, and other companies with storage, computing, and communications expertise.

Broadcom completed its acquisition of Sandburst Corporation on March 2, 2006. Broadcom described Sandburst as a fabless semiconductor company specializing in scalable packet switching and routing system-on-chip designs for enterprise core and metropolitan Ethernet networks.

What Happened to Advanced Switching

Advanced Switching did not become the dominant industry fabric. The broader market moved in other directions. Ethernet kept improving as a system and data-center fabric. PCI Express remained central for local I/O, GPUs, SSDs, accelerators, and endpoint expansion. InfiniBand grew in high-performance computing. Later, CXL built cache-coherent and memory-expansion semantics on top of PCI Express. Within telecom equipment, AdvancedTCA and other modular designs used several fabric options, and vendor-specific architectures continued.

That does not make Advanced Switching irrelevant. It is a useful signpost. Many of the problems AS tried to solve are still important: low-latency board-to-board traffic, fabric discovery, quality of service, redundancy, peer-to-peer transfers, hot-plug behavior, and efficient movement between processors, accelerators, storage, and network interfaces.

Modern Echoes

PCIe switches: modern systems use PCIe switches extensively for GPUs, NVMe drives, NICs, and accelerators, though usually under the mainstream PCIe model rather than AS.
CXL: Compute Express Link uses the PCIe physical layer to add cache coherency, memory expansion, and composable infrastructure features for CPUs, accelerators, and memory devices.
Ethernet fabrics: data centers and telecom systems now use high-speed Ethernet, RoCE, EVPN, and lossless or near-lossless designs for many fabric roles once considered for specialized interconnects, including some storage networking paths.
AdvancedTCA and modular systems: carrier-grade chassis continued to need high-availability switched fabrics, management planes, hot-swap behavior, and redundant paths.
AI servers: accelerator systems now expose the same old pressure at a new scale: how to connect many high-bandwidth devices with low latency, predictable behavior, and manageable power.

Lessons from Advanced Switching

Advanced Switching shows how hard it is for a fabric standard to win. Technical merit is not enough. A successful interconnect also needs broad silicon support, operating-system support, management tooling, volume economics, clear use cases, and timing that matches industry investment cycles. PCI Express had those conditions for local I/O. Ethernet had them for networking. Advanced Switching landed between worlds and never displaced the stronger ecosystems around it.

The lesson for current architectures is useful: before adopting a specialized fabric, ask whether the performance gain justifies the ecosystem cost. If an existing fabric can evolve to meet the need, the market often prefers the familiar path.

Planning Checklist

Use "Advanced Switching" carefully; in this context it means the historical PCI Express-based fabric specification, not generic switch features.
For modern PCIe fabrics, evaluate mainstream PCIe switching, retimers, cabling, topology, hot-plug support, and management tooling.
For memory and accelerator pooling, evaluate CXL rather than historical Advanced Switching.
For packet fabrics, compare Ethernet, InfiniBand, and proprietary accelerator interconnects based on latency, bandwidth, congestion handling, software stack, and deployment scale.
For carrier-grade modular systems, design high availability, management failover, serviceability, and fabric redundancy as first-order requirements.

Advanced Switching belongs to the history of PCI Express and modular communications equipment. It did not become the universal fabric its backers hoped for, but it anticipated a problem that has only become larger: systems need faster, more flexible, more reliable ways for many devices to communicate without turning every connection into a custom backplane.

Advanced Switching - Yenra