The h3llo.cloud infrastructure is currently distributed across five availability zones in Moscow, with plans to expand to six. The company operates within IXcellerate North and South data centers, with plans to migrate to its own facilities. Connectivity between key locations is provided by dark fiber links with a capacity of 400 Gbps, while the total network capacity reaches 800 Gbps.
The network architecture is based on the Dragonfly+ topology, utilizing a full mesh design between availability zones. Each zone is physically isolated with independent power, cooling, and network hardware. The company also maintains a facility in Alexandrov, which includes immersion cooling systems, though it is not used for commercial cloud clients.
The hardware stack consists of 11th-generation HPE servers equipped with NVMe and SSD storage. The company is currently procuring NVIDIA B300 GPUs for inference and data platform tasks, which are available under long-term commitments of at least one year. Future hardware upgrades will include 12th-generation HPE or 17th-generation Dell servers.
The cloud platform is built on proprietary software rather than third-party frameworks like OpenStack. The control plane and workload orchestration are based on Kubernetes, utilizing KubeVirt for virtualization, Linstor for software-defined storage, and Kube-OVN for networking. The company is pursuing FSTEC certification for its platform.
The company targets an SLA of 99.5% for its public cloud, with plans to reach 99.99% for virtual machines deployed across multiple availability zones. Service availability is calculated including network performance. Compensation for downtime is provided through discounts, starting at 10% for availability falling below the established thresholds.
