vSAN ESA Demonstrates Superior Performance Over Mainstream High-End Storage Arrays in Large Financial Enterprise Scenarios #
We are often asked by customers: “How does the performance of vSAN Express Storage Architecture (ESA) compare with existing storage arrays?” We wish there was a standard answer. However, due to the many variables present in data center environments, the most accurate reply is usually: “It depends on the specific case.”
Nevertheless, this is an important question worth exploring. In this article, we will provide a more detailed answer based on actual test results from a large customer evaluating the performance of vSAN ESA against a mainstream storage array.
Without further ado, how does vSAN ESA perform compared to the customer’s top-tier, capacity-equivalent traditional storage array? Here are the key highlights:
- In application-level tests, vSAN ESA delivered 20% higher IOPS performance than the traditional storage array, while maintaining sub-millisecond low latency.
- In synthetic I/O storage tests, vSAN ESA’s IOPS performance improved by up to 70%, still keeping sub-millisecond latency.
- Under hardware failure conditions, vSAN ESA’s latency dropped by 61%, showing stronger fault recovery performance.
- Beyond performance advantages, the customer’s internal evaluation showed that vSAN ESA’s total cost was 31% lower than their existing storage array while achieving higher performance!
Yes, the gap is indeed this significant. Next, we will further analyze the test results to help you better understand the meaning behind these numbers.
The Complexity and Challenges of Storage Performance Evaluation #
Besides data availability, delivering sufficient and consistent performance is a core responsibility of enterprise-grade storage solutions. However, measuring storage performance realistically and effectively—close to actual workloads—is not easy. For a long time, the storage industry has struggled with this issue—many vendors often promote exaggerated “hero numbers” (theoretical peak performance achievable under ideal conditions) to market their products.
Synthetic I/O generators do have their place in storage evaluation, but their results usually focus on total IOPS and do not truly reflect real application performance on different systems. Additionally, some synthetic tools may be deliberately configured to bypass known system bottlenecks, producing overly optimistic test results.
In this performance comparison, we deliberately avoided such one-sided or distorted tests. Fortunately, a large financial services customer allowed us to observe their testing of vSAN ESA against their existing top-tier storage array. The customer wanted clear insights into the performance differences, and we sought real, representative results.
Direct Comparative Analysis in a Real Environment #
Most workloads tested by the customer come from high-transaction-volume databases running mainly on SQL Server. These workloads have very high demands for low latency and performance consistency. During the evaluation of vSAN ESA versus mainstream high-end storage arrays, the customer particularly wanted results reflecting real production needs.
They focused on three key scenarios:
- High transaction pressure periods: e.g., during volatile market periods, can the system sustain large-scale transaction processing?
- Regular trading periods: the system’s stability and responsiveness under continuous load during normal market operations.
- Failure scenarios: system behavior during traditional storage array controller failures or vSAN host failures.
They also wanted to use traditional synthetic I/O tools to stress test both systems to understand their peak performance fully. Hence, the tests were divided into two main categories:
SQL Workload Test (Simulated Financial Transaction OLTP) #
This test simulated OLTP database workloads on 50 virtual machines running SQL Server, using a TPC-E-like benchmark. This benchmark measures overall performance of 50 VMs simulating a financial brokerage trading scenario. This test includes not only storage operations but also substantial complex non-storage computation tasks. Thus, overall performance is primarily limited by VM CPU and quantity.
Since its bottleneck is more on compute resources than storage, this is considered a “compute-constrained test.” The same approach was used to evaluate performance under failure conditions.
Synthetic Storage I/O Performance Test (Stress Test for Peak Performance) #
This test is a more common storage evaluation method, measuring the theoretical maximum performance of the entire storage system in IOPS, throughput, and average latency. It is a typical “storage-constrained test.” The test used the FIO tool as the synthetic I/O generator on both storage systems.
On the vSAN side, the HCI Bench tool (also based on FIO) was used to automatically deploy a suitable number of worker VMs. The test used 8KB block size with a 70/30 read/write ratio, matching the actual I/O pattern characteristics of the customer’s SQL business VMs.
Test Environment Configuration #
To ensure consistent and representative evaluation, the test used a 6-host vSAN ESA cluster. Each host roughly matched the vSAN ESA All-Flash 6-node ReadyNode configuration specification, including:
- 56 CPU cores
- 6 NVMe storage devices
- 768 GB memory
- 25GbE network interface
The environment ran vSAN version 8U3. All VMs applied a storage policy with fault tolerance level FTT=2 (RAID-6).
Test Results Analysis #
Let’s review the key test results and interpret their significance.
SQL Workload Test Results #
This test evaluates performance under high transaction pressure with a fixed number of business applications running on two different storage systems. Because this simulates financial OLTP transaction processing, the performance gap between storage systems is generally smaller than in synthetic I/O stress tests focused solely on storage.
Using the TPC-E-like benchmark on 50 VMs running OLTP applications:
- On the vSAN ESA cluster, the workload produced 150,000 IOPS with an average latency of 0.93 ms.
- On the comparison storage array, the same workload produced 125,000 IOPS with an average latency of 0.8 ms.
Compared to the traditional array, applications running on vSAN ESA gained 20% higher performance with nearly the same average latency, indicating higher application efficiency on vSAN.
Application Behavior Under Normal Load and Failure Scenarios #
When simulating a milder transaction rate on the vSAN cluster (typical daily load of ~35,000 IOPS), average latency dropped to 0.65 ms, showing better responsiveness under normal trading conditions.
In failure scenarios, results are instructive. When a single host failed in the vSAN cluster, application latency rose to 1.71 ms. When a controller failed in the traditional array, latency spiked to 4.33 ms. This means vSAN’s effective latency was 61% lower under similar failure conditions, demonstrating superior stability and fault recovery.
Synthetic Storage I/O Stress Test Results #
This test focuses on the theoretical maximum performance of storage systems and typically shows the biggest differences between architectures, as it nearly bypasses application-layer loads to test backend storage capability.
Using 8KB block size with 70/30 read/write:
- The 6-node vSAN ESA cluster achieved 858,000 IOPS with an average latency of 0.95 ms.
- The comparison storage array achieved only 500,000 IOPS with latency of 1.0 ms.
In other words, at similar latency levels, the vSAN ESA cluster’s overall IOPS improved by up to 70%, significantly outperforming the traditional array.
For critical business workloads requiring ultra-low latency and consistent performance, these results are clear guidance. vSAN ESA within a VCF environment offers superior performance to mainstream storage arrays, enabling more efficient application operation while maintaining sub-millisecond latency even in failure scenarios.
These test outcomes align with feedback from many customers since the launch of vSAN ESA: vSAN provides a clear performance advantage over their previous storage solutions.
Beyond Performance: Additional Value of vSAN #
The results clearly show that vSAN ESA easily outperforms traditional storage arrays. But vSAN’s advantages go far beyond just better IOPS, throughput, or latency metrics.
As a truly distributed architecture hyper-converged storage solution, vSAN naturally supports horizontal scalability. Whether building aggregated vSAN HCI clusters or disaggregated vSAN storage clusters, each new host adds linearly scalable storage capacity, compute power, and network bandwidth.
Most importantly, this scalability is predictable and proportional, greatly simplifying the technical and financial complexity of enterprise expansion, avoiding challenges faced by modular storage arrays in traditional three-tier architectures.
Of course, cost is also critical. Since VCF licenses authorize 1 TiB of vSAN capacity per core, vSAN’s actual per-TB cost is 31% lower than the customer’s existing storage array, while significantly boosting storage performance.
Conclusion #
vSAN is a high-performance storage solution that can deliver better performance for your applications. Built into the hypervisor and included with VCF licenses, such a cost-effective storage solution—what are you waiting for?