Hyperconverged Infrastructure (HCI) was designed to simplify IT operations by combining compute, storage, and networking into a single platform. While the promise of consolidation is appealing, especially from a cost-savings perspective, the reality of delivering high performance in an HCI environment is a formidable challenge.

In a three-tier architecture, a dedicated compute tier generates the work, a dedicated network handles communication between VMs, another dedicated network manages communication with storage, and specialized storage arrays read and write the data to deliver the results. In contrast, HCI clusters are tasked with doing it all. They host the virtual machines (VMs) performing the work, host the storage services, often as a VM, responding to the work, manage the storage media reading and writing the data, and provide networking services that deliver the results-all within the same infrastructure. This "double duty" within HCI can lead to significant performance bottlenecks, especially when workloads require both high throughput and sub-millisecond response times.

Achieving this level of performance requires more than a traditional HCI solution. It demands an integrated, efficient infrastructure that can properly distribute the conflicting demands of compute and storage across the cluster. This is where Ultraconverged Infrastructure (UCI) emerges as a better alternative by integrating virtualization, storage, and networking into a single codebase where each service is aware of the demands of the others and I/O can be more evenly distributed.

In UCI, storage is not treated as a separate VM but is integrated directly into the hypervisor. This design eliminates the overhead of running storage as a second-class citizen and ensures that storage I/O is prioritized alongside application workloads.

Integration also has a cascading effect. For example, if the solution has an inline deduplication feature, all services benefit from that feature. Network bandwidth and cache memory effectively increase in size following the deduplication ratio, further enhancing overall efficiency and performance.

Real-World Performance with UCI

As an example of UCI in action, VergeIO recently published a benchmark demonstrating the capabilities of UCI to handle demanding workloads. In this test, VergeOS was able to deliver:

• 1 Million+ Read IOPS using 64K blocks
• 480K Write IOPS using 64K blocks and 30 GB/s throughput

It is important to note that they used 64K blocks instead of the more benchmark-common 4K blocks, and that they achieved these results with only six standard servers (nodes) that would be common in almost any data center.

Conclusion

The difficulty of delivering high performance with HCI lies in its design-tasking the same infrastructure with doing all the work and delivering all the results. Traditional HCI architectures struggle to balance these demands, leading to resource contention, inconsistent performance, and scalability challenges.

UCI solves these problems by deeply integrating compute, storage, and networking into a unified platform. With its efficient architecture and intelligent resource allocation, UCI delivers the performance, scalability, and flexibility that modern workloads demand.

##