An enterprise has long relied on VMware vSAN to support its core business—the Omissa VDI desktops. However, following adjustments to VMware’s licensing policies, vSAN costs have surged, leading to a significant spike in the overall TCO. Given that the VDI environment is critical to the office day-to-day work, the enterprise aims to identify a potential vSAN alternative and achieve a production-level transition without impacting business continuity.
Impressed by the 61% vSAN-compared cost slashing of the joint solution of Arcfra AECP and Omnissa Horizon, the enterprise conducted thorough PoC on AECP under various business scenarios. The result revealed that for Omnissa VDI use cases, the total bandwidth of AECP with VMware ESXi reached 77% of NVMe disk performance and achieved 115% of vSAN performance after precise tuning.
The enterprise deployed both vSAN and Arcfra ACOS with VMware ESXi on the same three identical nodes and conducted performance tests in scenarios including VM creation, IOMeter, Diskspd, cross-disk file copy within VMs, and copying 3,000 small files.
| Hardware | Configuration |
|---|---|
| Node | ThinkSystem SR650 V3 |
| CPU | Xeon® Gold 5418Y |
| Memory | 256GB |
| Disk | S4620 SATA 960GB * 4 |
| Software | Editions |
|---|---|
| vCenter Server | 7.0.3 24322018 |
| ESXi | 7.0.3 23794027 |
| vSAN | 7.0 |
| Arcfra AECP | 6.2 |
| Diskspd | Performance testing |
| Test Method | 3P1V, VDI Windows VM:4CPU,16G MEM, 100G disk |
Initial results showed that AECP write performance was roughly comparable to vSAN, while read performance was significantly below expectations.
| Block Size | Access Mode (100% Random Read) | Mixed Workload (80% Read / 20% Seq) | Arcfra IOPS (Average) | VMware IOPS (Average) | Performance Differential (Arcfra/VMware) % |
|---|---|---|---|---|---|
| 4K | 100% Random Read | - | 22,963 | 42,315 | 54.27% |
| 4K | - | 80% Read / 20% Seq | 38,500 | 40,043 | 96.15% |
| 8K | 100% Random Read | - | 15,246 | 28,140 | 54.18% |
| 8K | - | 80% Read / 20% Seq | 24,800 | 28,695 | 86.43% |
| 32K | 100% Random Read | - | 7,150 | 11,850 | 60.34% |
| 32K | - | 80% Read / 20% Seq | 11,200 | 13,201 | 84.84% |
| 64K | 100% Random Read | - | 3,940 | 6,420 | 61.37% |
| 64K | - | 80% Read / 20% Seq | 6,400 | 8,462 | 75.63% |
| Test Scenario | Read/Write | Arcfra (MB/s) | VMware vSphere | Performance Ratio (Arcfra/VMware) % |
|---|---|---|---|---|
| SEQ1M Q8T1 | Read | 1139.01 | 1248.73 | 91.21% |
| (Sequential R/W, QD8, 1 Thread) | Write | 761.13 | 471.65 | 161.38% |
| SEQ1M Q1T1 | Read | 342.5 | 530.39 | 64.58% |
| (Sequential R/W, QD1, 1 Thread) | Write | 141.98 | 322.33 | 44.05% |
| RND4K Q32T1 | Read | 209.25 | 317.21 | 65.97% |
| (Random R/W, QD32, 1 Thread) | Write | 127.48 | 118 | 108.03% |
| RND4K Q1T1 | Read | 12.96 | 28.01 | 46.27% |
| (Random R/W, QD1, 1 Thread) | Write | 10 | 8.78 | 113.90% |
Following unsatisfactory results in initial testing, Arcfra technical team conducted a deep-dive investigation.
It was found that VMware VMs exhibited severe performance fluctuations during sequential read tests with an iodepth of 1. Preliminary analysis indicates that this issue is linked to the VMware vmxnet3 virtual network driver, suggesting a potential defect at the driver layer.
To address this, the enterprise and Arcfra technicians optimized the vNIC parameters directly within the SCVM. Specifically, the interrupt coalescing wait time was reduced from the default 250μs to 20μs.
This adjustment effectively mitigated the latency jitter caused by the coalescing policy, resulting in stable read performance – in the random read/write test on a 1GB file using Diskspd, the total bandwidth of AECP with VMware ESXi reached 77% of NVMe disk performance and achieved 115% of vSAN performance.
* Note: VMware vSAN and Arcfra AECP were both configured with SATA S4620. The tuning method was to change rx-usecs from 250 to 20.
Upon completion of the tuning process, the end user independently performed a performance re-evaluation. The findings verified the conclusions drawn from the second round of testing, with the AECP demonstrating stable performance that surpassed initial projections.
Arcfra AECP with VMware ESXi achieved 23% higher performance than vSAN.

* Note: PC server was configured with NVMe PM961, Arcfra AECP with SATA S4620, and VMware vSAN with SATA S4620.
A 5GB file was generated inside the VM and copied from drive C to drive D. The test revealed that Arcfra was 7 seconds (60%) faster.
A folder containing 3,000 files was generated inside the VM and copied from drive C to drive D. The test revealed that Arcfra was 8 seconds (60%) faster.

Testing and tuning results demonstrate that, with proper configuration and parameter optimization, Arcfra AECP read/write performance can match or exceed vSAN performance in the Omnissa VDI scenario, serving as a competitive alternative to VMware vSAN VDI performance.
Learn more about Arcfra x Omnissa joint solution for VDI and how it helps to save cost:
Arcfra Launches Joint VDI Solution with Omnissa, Slashing Costs by 61% Compared to VMware vSAN
Why Arcfra Is the Smarter Choice for Omnissa Horizon VDI
Best Practice Guide for Omnissa x Arcfra Solution | Free Download
Arcfra simplifies enterprise cloud infrastructure with a full-stack, software-defined platform built for the AI era. We deliver computing, storage, networking, security, Kubernetes, and more — all in one streamlined solution. Supporting VMs, containers, and AI workloads, Arcfra offers future-proof infrastructure trusted by enterprises across e-commerce, finance, and manufacturing. Arcfra is recognized by Gartner as a Representative Vendor in full-stack hyperconverged infrastructure. Learn more at www.arcfra.com.