To stop storage from becoming the bottleneck, match the storage system’s aggregate throughput to the bandwidth of the full I/O path: SSDs, PCIe connections, host adapters, network links, and switches. If any segment is narrower—or the fabric is oversubscribed—faster drives alone will not deliver faster application I/O.
What bandwidth matching means
Bandwidth matching is sizing each part of the storage path so it can carry the throughput the next part can produce. In a shared NVMe-over-Fabrics (NVMe-oF) system, that means comparing the aggregate capability of the SSDs with the host-facing network, switch fabric, and server connections—not just looking at a drive’s advertised speed.
Western Digital describes this goal for its OpenFlex Data24 4000 Series as “Bandwidth match between PCIe Gen4 SSDs (Storage) and I/O (Network) – No oversubscription.” The point is broader than that product: end-to-end throughput is limited by the narrowest or most congested part of the path.
How a mismatch creates an I/O bottleneck
Storage can outpace the network
NVMe SSDs can produce data faster than a traditional storage network can carry it. HPE Aruba notes that the bottleneck in a storage area network has shifted from storage media to the network. When the network cannot keep up, adding faster SSDs may increase the amount of data waiting to be sent without increasing application throughput.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems#1 Best Overall
- MEET THE NEXT GEN: Consider this a cheat code; Our Samsung 990 PRO Gen4 SSD helps you reach near max performance with lightning-fast speeds; Whether you’re a hardcore gamer or a tech guru, you’ll get power efficiency built for the final boss
- REACH THE NEXT LEVEL: Gen4 steps up with faster transfer speeds and high-performance bandwidth; With a more than 55% improvement in random performance compared to 980 PRO, it’s here for heavy computing and faster loading
- THE FASTEST SSD FROM THE WORLD'S FLASH MEMORY BRAND: The speed you need for any occasion; With read and write speeds up to 7450/6900 MB/s you’ll reach near max performance of PCIe 4.0 powering through for any use
- PLAY WITHOUT LIMITS: Give yourself some space with storage capacities from 1TB to 4TB; Sync all your saves and reign supreme in gaming, video editing, data analysis and more
- IT’S A POWER MOVE: Save the power for your performance; Get power efficiency all while experiencing up to 50% improved performance per watt over the 980 PRO; It makes every move more effective with less consumption
Oversubscription builds queues
Oversubscription occurs when the combined traffic that can arrive at a link exceeds that link’s capacity. The excess waits in a queue. Queues add latency; if they grow too large, packets can be dropped, prompting retransmissions and further slowing I/O.
The IETF’s BBR congestion-control draft describes the general principle: a flow’s sending rate should match the available bottleneck bandwidth, and its in-flight data should fit the path’s bandwidth-delay product (BDP). Sending faster than the bottleneck can deliver creates a queue; keeping more data in flight than the path can absorb can leave a standing queue that raises latency and loss risk.
Rank #2
- Ideal for high speed, low power storage
- Gen 4x4 NVMe PCle performance
- Up to 6,000MB/s read, 4,000MB/s write
- Includes Acronis cloning software
- 5-year limited warranty
Other parts of the path can be narrower
A network link is only one possible limit. Host NICs, PCIe connectivity, switch uplinks, cabling, metadata services, and storage software can also constrain results. For workloads dominated by small random operations or file lookups, metadata and IOPS behavior may matter more than headline sequential bandwidth.
How NVMe-oF transports storage traffic
NVMe-oF makes NVMe storage available to remote servers over a network, allowing multiple hosts to share flash rather than each relying only on locally attached drives. The transport affects latency, CPU use, and operational complexity.
Rank #3
- SPEED UP PROJECTS. Launch creator applications fast with uncompromising PCIe 4.0 read speeds up to 7,100MB/s,[2] (1TB and 2TB[1] models) and write speeds up to 6,700MB/s[2] (1TB[1]-4TB[1] models).
- CREATE AND STORE MORE. Make more room for your 4K videos and high-resolution images with capacities from 500GB[1] up to 4TB[1] on M.2 2280 built with our trusted 8th generation SANDISK BiCS QLC 3D CBA NAND.
- IT GOES WHERE YOU GO. With an all-new power efficient design, your drive delivers high performance with low power, giving you more time to be productive while on the go.
- UNCOMPROMISED RELIABILITY. With up to 1,200 TBW[3] (4TB[1] model) endurance rating, your drive is designed for creators.
- KEEP YOUR DRIVE UPDATED. Monitor your SSD’s performance and check for updates with the downloadable SANDISK Dashboard application.[5]
| Transport | Potential advantage | Trade-off |
|---|---|---|
| RoCE/RDMA | RDMA can place data directly in host memory and reduce CPU work; it is suited to low-latency access when the Ethernet fabric is engineered appropriately. | RoCE depends on a carefully configured, lossless Ethernet path. Packet loss and congestion require attention across the fabric. |
| TCP | Generally easier to deploy and compatible with broader network environments. | Loss can cause TCP retransmissions, which may reduce application performance. |
Neither transport removes the need to size the path. A low-latency protocol cannot compensate for insufficient aggregate link capacity, and a nominally high-speed fabric can still perform poorly if it is congested or lossy.
How to assess a storage and network design
- Estimate aggregate storage demand. Use the expected concurrent workload and the storage tier’s realistic aggregate throughput, not just the maximum speed of one SSD.
- Trace the complete data path. Check drive interfaces and PCIe generation, storage controllers, target and host NICs, switch ports and uplinks, and host connections.
- Calculate oversubscription at shared links. Compare the combined traffic that may converge on each uplink with that link’s usable bandwidth. Include simultaneous host activity and leave headroom for bursts and other traffic.
- Match the transport to the network you can operate. Choose RoCE when low-latency RDMA behavior is valuable and the team can engineer and monitor the lossless Ethernet fabric. Choose TCP where simpler deployment and broad compatibility matter more.
- Validate for the actual workload. Measure throughput, latency, packet loss, retransmissions, CPU use, and metadata performance with the intended host count, software, and traffic pattern. Vendor figures from a different configuration are not a reliable prediction.
What published examples do—and do not—show
Western Digital’s 2024 brief describes the OpenFlex Data24 4000 Series as a 2U NVMe-oF chassis with up to 24 U.2 NVMe SSDs, up to 368 TB of capacity, and 12 100GbE ports. It supports RoCE or TCP host connectivity. Those specifications illustrate one approach to providing substantial flash capacity and network bandwidth in a shared system; they do not establish the throughput every deployment will achieve.
Rank #4
- HUGE SPEED BOOST: Get random read/write speeds that are 40%/55% faster than 980 PRO; Experience up to 1400K/1550K IOPS, while sequential read/write speeds up to 7,450/6,900 MB/s reach near the max performance of PCIe 4.0*
- BREAKTHROUGH POWER EFFICIENCY: Use less power and get more performance; Enjoy up to 50% improved performance per watt over 980 PRO, plus optimal power efficiency with max PCIe 4.0 performance**
- SMART THERMAL CONTROL: Samsung's own nickel-coated controller delivers effective thermal control; With its slim size, 990 PRO is a perfect fit for desktops and laptops that meet the PCI-SIG D8 standard***
- THE CHAMPION MAKER: Up to 65% improvement in random performance enables faster loads for an ultimate gaming experience on PS5 and DirectStorage PC games****
- SAMSUNG MAGICIAN SOFTWARE: Get the most out of your SSD with Samsung Magician's advanced yet intuitive optimization tools; Monitor drive health, protect valuable data, and receive important updates for your 990 PRO
Harvard Edge gives a parallel-file-system example in which 20 storage servers, each providing 10 GB/s, yield 200 GB/s aggregate. That arithmetic describes aggregate server contribution; realizing it at clients also depends on the network fabric, client connections, and workload.
NVIDIA reports GPUDirect Storage gains on DGX platforms of 2.5× for deep-learning inference, 8× for data-analytics visualization, and 9× for video analytics. These are vendor-reported, workload-specific results, not general multipliers for other systems. Treat them as configuration-dependent claims and compare them only when the topology and software versions are relevant to your environment.
Recommended Free Tools
Best Value
- This product has been replaced by our latest generation. Please search for the SANDISK Optimus GX 7100 NVMe SSD
- HIGH-OCTANE GAMING. Experience speeds up to 7,250MB/s read and 6,900MB/s write (1-2TB models), with up to 35% faster performance than previous generation.
- PURPOSE-BUILT. Designed for serious on-the-go gamers, with a PCIe Gen4 interface and SANDISK’s next generation TLC 3D NAND.
- MORE TIME TO CLEAR THAT CHECKPOINT. Built with laptops and handheld gaming devices in mind, with up to 100% more power efficiency over the previous generation.
- DO MORE WITH DASHBOARD. Ensure your drive is optimized for prime performance with the downloadable WD_BLACK Dashboard (Windows only).
What to compare before choosing a system
- Capacity and media: drive count, supported U.2 or U.3 form factor, endurance, and usable capacity.
- Fabric bandwidth: total port speed, PCIe generation, host count, switch oversubscription, and cabling.
- Transport: RoCE/RDMA requirements versus TCP’s operational simplicity and retransmission behavior under loss.
- Latency and resilience: queueing, packet loss, retransmissions, redundancy, and failure domains.
- Scaling: whether aggregate data bandwidth grows with added storage servers, and whether metadata or control-plane services become limiting.
- Workload fit: sequential bulk transfers, random IOPS, AI training, analytics, databases, and shared virtualization stress different parts of the system.
For enterprise NVMe-oF systems, drive compatibility also matters. A consumer PCIe Gen4 NVMe SSD is not automatically a suitable replacement: enterprise U.2/U.3 form factor, endurance, firmware, and dual-port requirements can restrict the options.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
