K2 Cloud, a top-10 Russian cloud provider, operates in four data centers in Moscow and one in Saint Petersburg. The company develops its own platform using Python and open-source solutions like OVN. Due to high demand for dynamic infrastructure adjustments, the team frequently installs, moves, and reconfigures hardware across sites.
Between 2018 and 2019, the company faced significant operational challenges, including power overloads, cooling issues, and network instability caused by an aging InfiniBand infrastructure. These technical failures led to frequent server reboots and storage accessibility problems, necessitating a complete overhaul of their hardware management strategy.
The company established a dedicated infrastructure team to centralize hardware operations. This team introduced standardized rack configurations, migrated from InfiniBand to a pure Ethernet leaf-spine network architecture, and replaced legacy storage systems with distributed storage solutions that ensure data redundancy across multiple physical nodes.
New standards include optimized rack space utilization, standardized power distribution, and real-time energy monitoring using smart PDUs integrated with Zabbix and DCIM systems. These measures prevent power overloads and ensure that infrastructure scaling remains consistent across both large cloud deployments and smaller private installations.
