Validated data delivery performance with F5 BIG-IP and MinIO

AI workloads depend on fast, reliable data movement between storage and compute. When latency and jitter disrupt that flow, throughput drops, model training slows, and GPU resources are underutilized.

By sitting between compute and storage, F5 BIG-IP helps optimize traffic flow, improve connection efficiency, and stabilize throughput across distributed AI environments. This helps reduce the performance impact of latency and jitter, protects storage from bursty demand, and supports more consistent data delivery for training, fine-tuning, and RAG workloads. For architects designing scalable AI infrastructure, treating the storage data path as a managed network discipline, rather than a direct application-to-storage connection, helps maximize GPU utilization, improve pipeline reliability, and support scalable AI training and inference.

SecureIQLab validated testing of F5 VELOS and BIG-IP Local Traffic Manager in front of a MinIO cluster and found up to 3.3x higher S3-compatible storage throughput under real-world network conditions. The results show how F5 helps improve AI data delivery by optimizing the data path between storage and compute across distributed environments.

Key findings from the report

• In a low-latency enterprise SD-WAN scenario (10 ms latency, 2 ms jitter), F5 VELOS increased S3 throughput from 62.5 Gbps to 121.7 Gbps — a 95% performance improvement.
• In a low-latency enterprise broadband/VPN scenario (10 ms latency, 5 ms jitter), F5 VELOS increased S3 throughput from 46.8 Gbps to 119.6 Gbps — a 155% performance improvement.
• In a high-latency multicloud backbone scenario (75 ms latency, no jitter), F5 VELOS increased S3 throughput from 10.9 Gbps to 41.5 Gbps — a 281% performance improvement.
• In a high-latency, high-variability edge scenario (75 ms latency, 5 ms jitter), F5 VELOS increased S3 throughput from 9.6 Gbps to 40.1 Gbps — a 316% performance improvement.
• In a high-latency SD-WAN multicloud scenario (75 ms latency, 2 ms jitter), F5 VELOS increased S3 throughput from 9.5 Gbps to 41.2 Gbps — a 332% performance improvement.

Download the SecureIQLab report to review the test methodology, architecture, and detailed throughput results validated by:

What you’ll learn in this report

Implications show latency is the primary limiter of S3-compatible storage throughput

• Latency between the S3-compatible storage client and the storage nodes affected throughput much more than expected.
  • The baseline results showed throughput at 10ms latency was reduced to ~60% of the 0ms latency baseline throughput.
  • The baseline results showed throughput at 75ms latency was reduced to ~8.25% of the 0ms latency baseline throughput.
• Latency had a greater impact on throughput between the S3 client and storage nodes than jitter. F5 expected jitter to have a greater relative impact on the throughput vs. latency.
• An F5 VELOS chassis with BIG-IP LTM acting as a proxy optimizes traffic between S3-compatible storage clients and storage nodes, increasing throughput on real-world networks at much higher rates than expected at the outset of the validation.

_{Note: Performance results are based on testing conducted by F5 in a controlled lab environment and validated by SecureIQLab. Results may vary depending on workload characteristics, infrastructure configuration, and network conditions.
Report originally published by SecureIQLab on April 2, 2026.}