In the race to make cloud infrastructure faster, smarter, and more efficient, one truth stands out: stateful networking is the next frontier. As applications demand richer services and tighter security, cloud providers must manage millions of active connections in real time without sacrificing speed or reliability.
That’s where SONiC-DASH and Xsight Labs come in. Together, we’re transforming how the world’s largest clouds move data.
What Is SONiC-DASH?
SONiC-DASH, short for Disaggregated API for SONiC Hosts, is an open-specification initiative under The Linux Foundation, originally created by Microsoft. It defines a new, standardized way to deliver stateful cloud networking services, such as firewalls, connection tracking, and load balancing.
The goal is simple but revolutionary: bring openness and interoperability to the parts of cloud networking that have traditionally been locked behind proprietary APIs or specialized hardware.
Just as SONiC revolutionized stateless switching in data centers, DASH is doing the same for stateful host networking by enabling cloud providers to choose from a wide ecosystem of optimized, cost-efficient implementations.
The Challenge: Stateful Networking at Scale
Every cloud connection tells a story, from the first handshake to the final packet. Managing those millions (or billions) of stories in real time is a massive challenge.
A DASH system with 800G throughput must handle:
- 12 million new connections per second
- Over 120 million active flows, without dropping a single packet
That’s the bar set by the DASH Hero Test, the ultimate benchmark for stateful cloud networking performance.
Until now, meeting that bar required scaling across multiple DPUs or specialized appliances, driving up cost and power consumption. Xsight has rewritten that equation.
The Breakthrough: DASH at 800G on a Single, and Open, Chip
With our E1, the first and only 800G Data Processing Unit (DPU) on the market, we have achieved what many thought impossible: meeting the full DASH Hero 800 benchmark requirements on a single device. When tested by an independent test lab, Xsight DASH exceeded the requirement, handling 14.3 million connections per second without a single dropped packet.
In this configuration, every inbound and outbound packet flows through the E1’s twin 400G Ethernet interfaces, while maintaining state for over 120 million Transmission Control Protocol (TCP) and User Datagram Protocol (UDP) flows.
This is more than just a performance number. It’s proof that open, disaggregated, stateful networking can run at hyperscale speeds efficiently, reliably, and cost-effectively.
Inside the Xvantage[BG1] [EV2] [JK3]
Xsight’s E1 DPU is a purpose-built powerhouse, combining:
- 64 Arm Neoverse N2 cores for high packet-rate processing
- 4 DDR5-5200 memory channels for over 1 Tbps of bandwidth
- 800GE of Ethernet connectivity
- And a fully standard Data Plane Development Kit (DPDK) based software model
What does that mean?
1. Smarter Use of Compute
The packets for each flow are pinned to a CPU core and processed in a simple run to completion programming model, enabling linear scaling with cores. Every 200ms, dedicated “aging cores” clean up timed-out connections, ensuring memory stays lean and performance stays consistent even under massive workloads.
2. Memory That Keeps Up
The E1’s DDR5 memory subsystem handles billions of random memory operations per second, while intelligent prefetching hides latency. In other words, the DPU keeps executing even while it’s waiting for data.
3. Offloads That Save Power
While fully software-driven, Xsight DASH selectively uses hardware offloads for parsing, lookups, hashing and encryption/decryption. This can be used to free up 20% or more of CPU resources, translating to lower power, increased headroom, and additional revenue generating capacity.
4. Resilience Without Waste
Xsight’s innovative high availability synchronization reduces bandwidth overhead to just 2%, compared to 17% in typical implementations. It supports instantaneous failover, as well as live migration of 128 million flows in under 30 seconds, delivering seamless upgrades and zero-downtime reliability.
See It in Action: The Xsight DASH-board
Performance isn’t just about speed, it’s about visibility. The Xsight DASH-board, a real-time text user interface, lets operators monitor performance per port, per core, or across the system. Through open Google Remote Procedure Call Application Programming Interface (gRPC APIs), the DASH-board also integrates easily into existing monitoring tools, giving operators instant insights into load, throughput, and health.
Xsight’s DASH implementation represents the future of open cloud networking:
- Faster scaling without hardware lock-in
- Lower operational costs through software-driven efficiency
- Higher reliability with next-gen HA and live migration
- Openness and flexibility through SONiC integration
Simply put, Xsight’s E1 DPU brings together the performance of custom silicon and the agility of open software.
By proving that a single DPU can run the full DASH workload at 800G, Xsight Labs has set a new benchmark for cloud-scale performance. It’s not just faster networking; it’s a blueprint for how open innovation can power the next generation of cloud infrastructure.
Welcome to your high performance and disaggregated future.