As part of our virtualization strategy, we evaluated alternatives to VMware with the goal of selecting platforms based on workload requirements rather than finding a single replacement for every use case.
For the majority of our workloads, we selected Oracle Linux KVM as the virtualization platform.
However, our GPU workloads introduced an additional requirement: vGPU support. At the time of our evaluation, the required vGPU capabilities were not officially supported on the Oracle KVM platform we were considering.
This led us to evaluate Microsoft Hyper-V as an alternative for GPU workloads.
Why Hyper-V?
Hyper-V provided the virtualization capabilities required for our GPU platform while also offering enterprise features such as:
- Failover clustering
- Centralized management through SCVMM
- Shared storage integration
- Virtual networking
- VM lifecycle management
Rather than treating Hyper-V as a replacement for our entire virtualization environment, we chose it specifically for workloads where GPU virtualization was an important requirement.
High-Level Architecture
The Hyper-V platform follows a straightforward architecture:
┌──────────────────────┐
│ SCVMM │
│ Central Management │
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ Hyper-V Cluster │
│ Hyper-V Hosts │
└──────────┬───────────┘
│
│ SMB
▼
┌──────────────────────┐
│ Shared Storage │
└──────────────────────┘
SCVMM provides centralized management and orchestration of the Hyper-V environment.
The Hyper-V cluster provides the compute layer and hosts the GPU-enabled virtual machines.
SMB shared storage provides centralized VM storage accessible by the Hyper-V hosts.
An important design point is that SCVMM operates as the management plane and is not in the VM storage data path. Hyper-V hosts communicate directly with shared storage.
A Workload-Based Strategy
The key outcome of this evaluation was a shift from asking:
“What single platform should replace VMware?”
to:
“Which platform best meets the requirements of each workload?”
Our resulting strategy is:
Workloads
│
┌─────────┴─────────┐
│ │
▼ ▼
General Workloads GPU Workloads
│ │
▼ ▼
Oracle Linux KVM Hyper-V
For most workloads, Oracle Linux KVM meets our virtualization requirements. For GPU workloads requiring capabilities that were not available on our selected KVM platform at the time of evaluation, Hyper-V provides an alternative.
The broader lesson is simple: a VMware replacement does not necessarily have to be a single platform. Workload requirements should drive the virtualization architecture.
Techbrainblog 