Our privately owned and operated Tier-3 data center in Athens provides our research projects the ability and flexibility to provide on-demand scalable and high performance computing, network and storage resources in a fault-tolerance embedded architecture, HPC compliant, Distributed Computing and Cloud technologies. The data center can support any required testing and experimentation research activity, and validation activities in the area of Cloud computing and also Deep Learning technologies.
Facility Key Features
- SECURE PHYSICAL ACCESS CONTROLS
- REDUNDANT POWER SYSTEMS
- AUTOMATED FIRE SUPPRESSION SYSTEM
- FAULT-TOLERANT HIGH-SPEED WAN CONNECTIVITY
- AUTOMATION-DRIVEN COOLING SYSTEMS
- QUICKLY EXPANDABLE TO MEET NEW WORKLOADS
- FAULT TOLERANT BY DESIGN (AT LEAST N+1)
Compute, Network and Storage
Our bare-metal servers consist of Supermicro’s latest generation compute units. Utilizing the Intel Xeon Gold series processors in a Dual-CPU configuration per node, multi-gigabit link aggregates and NVMe storage, our server farm is able to handle our moderately complex and heavy workloads efficiently. Supermicro open switches provide fault tolerant, multi-terrabit switching capacity which interconnects the compute and storage nodes. More specifically, it consists of 10 x Intel Xeon Gold 5218 processors, 640GB RAM, 20TB NVMe storage and multi-gigabit node interconnections using two Supermicro open switches.
Our testbed is also equipped with high performance GPU for Deep Learning workloads
Expandability and Upgrade Paths
The architectural approach implemented for our data center infrastructure is to be able to provide drop-in plug and play hardware upgrades which expand our capabilities rapidly.
Our GPU upgrade path provides immense and immediate resources for demanding workloads in a scalable fashion. Expandable up to 24.6 TerraFlops of double-precision GPU compute capacity per 1 Rack Unit, in a total GPU core density of over 15,000 GPU cores per 1 Rack Unit of space. On mainstream research workloads, these resources can provide performance ratios of 1:135 comparing a regular server-class CPU.
Following our architectural approach to provide drop-in upgrades to all of our resources, our already high-performance networking technologies stack allows us to provide plug-and-play upgrades to allow for even higher performance with Infiniband and multi-100Gbit Ethernet inter-node connectivity.
Our storage tiers provide primary workloads with NVMe SSDs and high performance SAS SSDs for secondary workloads. All of our drives are redundant N+1 across our servers. Expandability in both tiers is plug-and-play to extend our capacity accordingly to a maximum of 100TB of redundant high performance storage.
QuComm EaaS Cloud Platform
QuComm operates a privately owned cloud based on comprehensive deployments of OpenStack and Kubernetes to manage bare-metal resources into virtual machines and containers, respectively. Our Everything-as-a-service (EaaS) philosophy is to be able to provide a high quality as-a-service approach to software and hardware resources. The entire architecture embeds scalability, high performance and security across all resources and tenants. Storage is also managed via OpenStack to provide several tiers of high capacity and performance driven storage pools.
Through orchestration and a best-of-breed DevOps approach, our platform is able to handle a multitude of workloads across Neural Networks, Deep Learning, Big Data and other areas of distributed and cloud computing which favors our areas of research and development.