.png){kind=logo}
RocketStor 4243AS: Disaggregated NVMe-oF Storage for 100GbE Enterprise Architectures
Dual 100GbE Wire-Speed Performance Validated by Western Digital's Open Composable Compatibility Lab (OCCL)
The HighPoint RocketStor 4243AS (RS4243AS) is a 2U, 24-bay Disaggregated NVMe-over-Fabrics
(NVMe-oF) Storage Platform engineered for IT solution providers, system integrators (SIs), and enterprise cloud builders. By liberating high-performance flash from isolated server chassis, the RS4243AS consolidates up to 24 U.2/U.3 enterprise NVMe SSDs (up to near 3-PB per 2U) into a centralized, composable storage target.
Operating across standard NVMe/RoCE v2 (RDMA) and NVMe/TCP fabrics, the RS4243AS acts as a high-density hardware bridge. It enables distributed host clusters to share remote flash pools with just ~10 microseconds of network transport overhead—combining the pooled efficiency of shared storage with the performance of direct-attached NVMe PCIe drives.
Proven Performance: Western Digital OCCL Lab Validation
The Western Digital Open Composable Compatibility Lab (OCCL) evaluated the RocketStor 4243AS populated with Western Digital Ultrastar DC SN650 15.36TB enterprise SSDs. The test results verify that the RocketStor 4243AS operates at the theoretical boundary of dual 100GbE network fabrics:
Performance Highlights
The OCCL test report validates that the RocketStor 4243AS achieves line-rate throughput and maintains local-NVMe response quality across a 100GbE fabric:
1. Near-Local Latency & Zero Fabric Penalty
Sub-Microsecond Transport Delta: Accessing storage across the network adds only ~10 µs of transport overhead over local DAS (10.01 µs delta for reads at 99.31 µs vs. 89.3 µs; 10.49 µs delta for writes at 21.29 µs vs. 10.8 µs).
True Local NVMe Responsiveness: Hardware-offloaded protocol translation preserves single-to low-microsecond boundaries, allowing remote pooled flash to match bare-metal direct-attached drive response times.
2. Line-Rate Saturation & High-Density IOPS
Dual 100GbE Saturation: Reaches 23,402 MB/s (23.4 GB/s) sequential read, achieving 99.5% line-rate efficiency of the physical dual 100GbE network boundary, alongside 18,103 MB/s (18.1 GB/s) sequential write.
Massive Parallel Scale: Delivers up to 4.08M 4K Random Read IOPS and 3.97M Mixed (70/30) IOPS, confirming that the platform handles heavy concurrent I/O queues without fabric or controller saturation.
3. Response Quality & SLA Predictability
Consistent Tail Latency: The QD1 exceedance profile tracks the local direct-attached baseline smoothly up to 6-nines (99.9999), with 4-nines (99.99) write latency remaining tightly bounded at 29.31 µs.
Deterministic Execution: Random IOPS Coefficients of Variation (CoV) stay between 0.030% and 0.077%, proving steady-state delivery and eliminating jitter or noisy-neighbor interference under sustained enterprise workloads.
4. Scope & Evaluation Boundaries
Focus of the Report: Validates aggregate throughput, raw line-rate saturation, and network transport latency using a standardized multi-drive benchmark configuration.
Not Explicitly Benchmarked: Multi-tenant host isolation boundaries across up to 120 initiators, environmental/thermal airflow profiles, and long-term NAND endurance wear metrics were outside the scope of this specific performance document.
Test Description | Local x1 | RocketStor 4243AS x16 |
|---|---|---|
Read Bandwidth (128KB) MB/s | 7,107 | 23,402 |
Write Bandwidth (128KB) MB/s | 3,003 | 18,103 |
Random Read (4KB)K IOPS | 983 | 4,079 |
Random Write (4KB) K IOPS | 110 | 1,880 |
Random Mixed (4KB) K IOPS | 278 | 3,966 |
4K Random Write Latency (μs) | 10.8 | 21.29 |
4K Random Read Latency (μs) | 89.3 | 99.31 |
4K Random Write 4-9s μs | 20.9 | 29,312 |
4K Random Read 4-9s μs | 213.3 | 305,830 |
Need More Details?
If you would like to receive the complete Validation and Performance Report, please submit the following: Report Request
Target Workloads & Enterprise Application Scenarios
Equipped with 24 independent NVMe bays, dual 100GbE connectivity, and support for NVMe/TCP and NVMe/RoCE v2, the RocketStor 4243AS enables organizations to separate storage from compute and scale NVMe capacity independently.
Cloud & Managed Service Providers (CSPs / MSPs)
The Challenge: Multi-tenant virtualization environments suffer from "noisy neighbor" syndrome, where heavy I/O from one tenant causes latency spikes and service-level agreement (SLA) breaches for others.
The RocketStor 4243AS Advantage: Delivers 4,079,000 Random Read IOPS and 3,966,000 Mixed Workload (70/30) IOPS, supporting high tenant density per rack unit. Deterministic 29.31 µs write tail latency (99.99) ensures predictable quality of service across all client instances under peak concurrent load. Native Redfish and OpenBMC support simplifies multi-tenant namespace partitioning and out-of-band billing telemetry.
AI Model Training Staging & HPC Checkpointing
The Challenge: GPU nodes (NVIDIA HGX / workstations) stall when storage pipelines cannot deliver training batches at line rate, while periodic checkpoint saves pause active model training loops.
The RocketAStor4243AS Advantage: Delivers 23.4 GB/s of continuous streaming read throughput, saturating dual 100GbE RoCE v2 links to prevent GPU compute starvation. Provides 18.1 GB/s sustained sequential write throughput, reducing checkpoint wait times so GPU clusters resume training cycles rapidly. Direct zero-copy RDMA data paths over RoCE v2 transfer data into GPU memory without host CPU buffer copies.
High-Throughput Media & Entertainment (M&E) Workflows
The Challenge: Collaborative 4K, 8K, and uncompressed multi-stream post-production pipelines demand high sequential throughput across multiple editing stations, while legacy SANs add administrative complexity.
The RocketStor 4243AS Advantage: 23.4 GB/s aggregate bandwidth supports concurrent editing of dozens of uncompressed 4K/8K ProRes and RAW streams without dropped frames. Deployable over standard, cost-effective 100GbE Ethernet switching infrastructures, avoiding the capital expense and maintenance overhead of Fibre Channel SAN fabrics. High-density 2U chassis accommodates up to 368TB+ of flash, providing a compact, centralized scratch disk for collaborative production facilities.
Post-VMware Migration / Software-Defined Storage (SDS) Enterprise Clusters
The Challenge: IT organizations migrating away from VMware to alternative virtualization stacks (such as Proxmox VE, Microsoft Hyper-V, KVM, or OpenStack) need high-performance, non-proprietary shared storage pools.
The RocketStor 4243AS Advantage: Standardized NVMe/TCP and RoCE v2 protocols integrate with Linux kernel target drivers and SDS platforms (including OSNexus QuantaStor & Ceph) without proprietary hypervisor software licensing. Adds only 10 µs of latency over local PCIe, enabling hypervisor clusters to access shared storage pools with the responsiveness of local server SSDs. Consolidates stranded server storage into a single high-density target, lowering rack footprint, power consumption, and ongoing management overhead.
The Right-Sized Storage Architecture: RocketStor 4243AS vs. WDC OpenFlex Data24
Point of Comparison | WDC OpenFlex Data 24 4200 | RocketStor 4243AS |
|---|---|---|
Silicon & Controller Topology | Multi-Adapter topology (3-6 RapidFlex ASICs) | Single integrated RapidFlex C2000 Adapter |
Network Fabric Interface | Up to 12x 100GbE QSFP28 Ports | Dual 100GbE QSFP28 Ports (200GbE Aggregate) |
Internal Switching Fabric | Multi-Bus module routing | HighPoint Rocket 1528 PCIe Gen4 Switch Fabric |
Sequential Read Throughput | Multi-Port aggregated (70+GB/s) | 23.4GB/s (99.5% Wire Saturation) |
Random Read IOPs | Scaled across multi-controller mesh | 4,079,000 IOPs (4.08M) |
Chassis Power & Thermals | Deep Chassis with dual 2000W PSUs | 2U active air-cooled chassis with redundant 1600W PSUs |
Solution Provider Value & Positioning
Architectural Simplicity: Eliminates inter-controller latency and the complexity of multi-ASIC management.
100GbE Alignment: Integrates directly into standard Top-of-Rack switches without leaf-spine rewiring
Non-Blocking Flow: Routes 24 drive lanes into the fabric path without PCIe credit stalls
100% Wire Saturation: Extracts the full line-rate over dual 100GbE connections
High IOPs Density: Exceeds 2M IOPs per 100GbE port with 10 μs latency overhead
Lower OpEx: Lower chassis heat dissipation protects SSD endurance and reduces facility PUE.
Technical Specifications:
Form Factor: 2U 19-inch rackmount chassis
Drive Bays: 24× Hot-Swappable 2.5-inch U.2/U.3 NVMe PCIe Gen4 SSD bays
Network Interfaces: Dual 100GbE QSFP28 ports (200Gbps aggregate bandwidth)
Fabric Protocols: NVMe over Fabrics via NVMe/RoCE v2 (RDMA) and NVMe/TCP
Fabric Controller: Western Digital RapidFlex™ C2000 NVMe-oF Controller
Internal Switching: Integrated HighPoint Rocket 1528D PCIe Gen4 switch fabric
Host Initiator Support: Up to 120 independent host connections
Out-of-Band Management: Dedicated 1GbE RJ45 management port, ARM OpenBMC, RESTful Redfish APIs
Power Supply: 1+1 Redundant 1600W Platinum-certified power supplies
_edited.png)