Modern data center architectures are undergoing a fundamental transformation. As AI workloads, cloud applications, and data-intensive services scale exponentially, traditional infrastructure faces severe bottlenecks driven by host CPU overhead, high power consumption, and rigid fixed-function hardware. To break through these limitations, we have engineered Xsight Accelerate, a suite of compute and networking solutions designed to increase ROI and decrease time to market.
We are launching the first two solutions as part of the Xsight Accelerate Program: Storage and the Network Packet Broker (NPB). Combining high-performance silicon, dedicated hardware offloads, and software-defined programmability, these new turnkey architectures deliver line-rate performance while reclaiming compute power back for host workloads.
Accelerate Storage Solutions: Line-Rate Efficiency with E1 DPUs
Traditional NVMe-over-Fabrics (NVMe-oF) architectures often rely heavily on host CPU processing, consuming valuable compute cycles for storage transport and data protection services. Built around the 800 Gbps E1 DPU architecture, Xsight’s storage solutions break down these bottlenecks by offloading demanding storage workloads directly to dedicated hardware.
Key Features and Capabilities:
- Wire-Speed Offloads & Security: The E1 DPU provides line-rate hardware block encryption using AES-XTS alongside native, low-overhead NVMe-oF acceleration across standard NVMe/TCP and RoCEv2 transport protocols.
- On-Chip Advanced Storage Services: Featuring a programmable run-to-completion engine, the E1 handles active replication, high-availability data protection, snapshots, cloning, and background erasure coding without host intervention.
- Flexible Core Deployment Architectures: Supports up to 8 directly attached PCIe Gen5 x4 NVMe SSDs in server target mode.
- RAM-Disk Emulation: Uses RAM-backed virtual NVMe devices to enable low-latency storage testing and rapid prototyping without requiring physical SSD hardware.
- SATA SSD Expansion: Scales up to 16 SATA SSDs via cost-effective PCIe-to-SATA expansion cards for capacity-centric workloads.
- Versatile Client Deployment Models: Offers flexibility across traditional host-based initiators, DPU-based NVMe emulation, and full DPU application offload where both the application and storage initiator run directly on the E1 DPU—minimizing host CPU involvement.
- Open Ecosystem Integration: Integrates into cloud infrastructures using standard Linux netdev interfaces, VirtIO layers, and high-performance SPDK user-space drivers.
NPB: Wire-Speed Tunnel Handling on X2 Programmable Switch
Network Packet Brokers are critical for providing total visibility across complex enterprise and cloud networks. However, legacy fixed-function NPB appliances struggle to adapt to evolving protocol formats without sacrificing line-rate speed. Powered by the X2 12.8T programmable switch, the NPB solution delivers a fully programmable, software-defined data plane capable of zero-packet-loss execution.
Core Capabilities & Advanced Tunnel Handling:
- Fully Programmable Pipeline: Uses a software-defined Parser and MAP engine pipeline compiled via the XDE toolchain and XASM (Xsight Assembly) microcode, ensuring field-upgradability without hardware redesigns.
- High-Density Performance: Supports up to 12.8 Tbps aggregate throughput across versatile port configurations, including 16x OSFP 800G or 32x QSFP ports.
- Comprehensive Core Functions: Combines multi-speed aggregation, L2–L7 filtering via dual TCAM banks, packet slicing/truncation, hardware timestamping, flow-key deduplication, and multicast replication.
- Line-Rate Tunnel Encap & Decap: Implements hardware-driven tunnel encapsulation and decapsulation for a wide array of networking protocols with proprietary headers:
- VXLAN & GRE: Strips or pushes outer ETH/IP/UDP/VXLAN and GRE headers to expose or encapsulate inner traffic.
- ERSPAN (Type II/III): Strips outer transport layers to feed mirrored analysis traffic directly to monitoring tools.
- MPLS & GTP-U: Strips label stacks or outer GTP-U headers for telecom and enterprise data stream inspection.
Transforming Next-Generation Infrastructure
Whether offloading high-throughput storage pipelines or processing complex network telemetry, the Xsight Accelerate program gives infrastructure architects the agility of software paired with the uncompromised performance of purpose-built silicon.
To explore technical documentation, solution briefs, and starter development kits contact our team below.