The SPECvirt Datacenter 2021 Benchmark - An Answer to the Datacenter Architect’s Dilemma

By Travis Hindley and Lisa Roderick, SPEC Virtualization Committee

An IT architect designing a virtualization-focused data center faces a variety of crucial choices. The following are just some of the complexities an IT architect must navigate to create a robust and efficient virtualized data center.

• Selecting the right virtualization platform is paramount. Options including VMware vSphere, RedHat OpenShift Virtualization, or an open source model based on KVM each offers distinct licensing costs, feature sets, support availability, and integration strengths.
• Storage design is a strategic decision. For example, utilizing software-defined storage (SDS) allows for flexible allocation of resources but requires expertise in managing virtualized storage pools.
• Network design is critical. Implementing network segmentation ensures secure isolation between virtual machines, while the choice of network fabric (e.g. Ethernet or InfiniBand) impacts virtual machine communication speed and cost.
• The architect’s server hardware specifications matter. Balancing core count, memory capacity, and storage options with workload demands ensures optimal performance and avoids resource bottlenecks.

All this complexity was on the mind of Kelly, an IT architect at a cloud services company. After a decade of hosting all IT infrastructure in the public cloud, Kelly was tasked with building a new on-premises virtualized datacenter that could offer the flexibility required to meet evolving data challenges for years to come. Kelly knew that a software defined datacenter (SDDC) has fundamentally changed how modern IT is deployed but that it requires careful analysis of the benefits and the risks. Fortunately, Kelly was able to use the SPECvirt Datacenter 2021 benchmark, an industry standard tool, to help work though all the complex analysis and decision making.

The Team and the Benchmark

Kelly and the team had already identified their workload requirements and SLAs. As part of the day-to-day operations, several organizations in Kelly's company require VMs to be deployed on demand. For example, DevOps frequently requires fresh VMs for their test cycles during their business applications implementation and their integration of supply chain instances due to a recent acquisition. The training organization requires their pre-configured VMs to be deployed every Sunday night to prepare for Monday's sessions. Their dynamic datacenter virtual environment requires failover, and they expect their VMs to load balance and migrate between hosts seamlessly. The proposed solution would also need to support long-running, CPU-intensive jobs, such as financial reporting, without affecting online activity.

In designing a solution, some team members were interested in Hyperconverged Infrastructure (HCI) as the latest technology for high-performing datacenters. Others were concerned that HCI might negatively affect performance of big data analytics and their investigations into ML/AI. They were also concerned about being tied to one vendor. Kelly decided that a proof-of-concept would help them evaluate the performance of HCI and non-HCI virtualized datacenter environments. By running the SPECvirt Datacenter 2021 benchmark, they would be able to objectively compare and fairly assess the performance of their infrastructure under real-world workload scenarios. Since SPEC is an industry-standard consortium, the benchmark is vendor-neutral, ensuring fairness and objectivity in performance evaluations. This would allow the team to make informed decisions without bias towards specific vendors or products.

Since the SPECvirt Datacenter 2021 benchmark models typical, modern-day usage of virtualized infrastructure, Kelly used it to test and compare the performance of operations in HCI and non-HCI configurations. This would enable cost savings by avoiding over-provisioning or the under-utilization of resources. The benchmark is flexible in that it offers various workloads that allowed the team to choose workloads similar to their workload profile. It allowed the team to run a subset of the available workloads that were most interesting to them; for example, they could test with only the department workloads if they wanted to measure the performance impact and speed of VM deployment from a template. The benchmark also offered the team scripting hooks that could spawn lower-level performance data collection, such as operating system and hypervisor performance data collectors.

Another benefit of the SPECvirt Datacenter 2021 benchmark is its ease of use. Kelly appreciated that the team didn’t need to manage the error-prone nature of manual testing. They didn’t need to create a VM, install the operating system into it, adjust specific OS tuning settings, install workload applications, or generate workload data. The SPECvirt Datacenter 2021 pre-built appliance contains the required components for the controller, workload driver clients, and workload VMs necessary for the benchmark’s operation. The software is pre-loaded and pre-configured, which minimized Kelly's need to intervene, reduced implementation time and effort, and improved accuracy.

The benchmark provisions increasing numbers of virtual machines and workload intensities, which allowed the team to test the scalability of their proposed virtualized environment. It helped them understand how a particular infrastructure configuration performed as demand increased. They could analyze the benchmark results to identify bottlenecks and inefficiencies in their proposed infrastructure configuration. And they could re-run the benchmark iteratively in HCI and non-HCI configurations using their optimizations and compare the results against previous runs to validate their effectiveness.

Conclusion

Running the SPECvirt Datacenter 2021 benchmark in HCI and non-HCI configurations enabled Kelly and the team to carry out their due diligence. The benchmark provided the hard data they needed in their analysis to support their recommendations, so they could make informed decisions as they architected their future datacenter.

[Back to SPEC blog]