Shared HPC Power, Fully Separated Data: One Platform for Many Institutions

Several universities needed to share one HPC and AI platform without sharing data. Dayo-Tech authored the specification and built a multi-tenant environment with a petabyte of storage, hard data isolation, and independent identity per institution.

The Decision

Collaboration between research institutions runs into a structural problem: they want to pool computing power, but none will pool its data. The Customer, a multi-institution academic research initiative, needed several universities to work on one HPC and AI platform while keeping each institution's data completely separate.

Conventional shared environments don't resolve that contradiction cleanly. Building a separate cluster per institution would waste resources and fragment the research effort; a naively shared one would violate the data boundaries each university depends on. The Customer needed infrastructure that was genuinely shared and genuinely separated at the same time — a requirement that drove the entire project.

The Architecture

Dayo-Tech's role began before any hardware existed. The company authored the complete tender specification for the cluster, defining the compute servers and one petabyte of storage. Starting at the specification stage, rather than executing someone else's design, meant the isolation and identity requirements shaped the architecture from first principles.

The design centered on multi-tenancy with hard boundaries. Centralized identity management let each university be governed independently, and granular permissions kept data fully partitioned between institutions. Containerized microservice applications were deployed with Docker Engine and Docker Compose, giving the platform a modular way to run workloads. Identity and separation were treated as core architectural decisions rather than controls added afterward, which is what allows one physical platform to serve many institutions without exposing any of them to the others.

The Platform

The platform combines GPU servers and HPC compute with one petabyte of shared storage, supporting deep learning and prediction workloads. On the surface it operates as a single system with a shared compute and storage pool. Underneath, it behaves as a set of separated environments, each governed by its own identity management and bounded by granular permissions.

That dual nature is the point: researchers across several institutions draw on the same underlying infrastructure, while each institution's data stays its own. For a multi-institution initiative, that is what makes shared research infrastructure viable at all.

Technology Stack

• Compute / GPU: GPU servers; HPC compute servers

• Storage: 1 PB storage system

• Virtualization / Containers: Docker Engine; Docker Compose

• Security / Identity: Centralized identity management; granular permissions and enforced data isolation

  
  

The Impact

A single multi-tenant HPC and AI environment now serves several institutions on shared GPU compute and a petabyte of storage, with hard data separation and independent identity management per university. The alternative, duplicate, siloed infrastructure operated separately by each institution, was avoided entirely. Described qualitatively, since no adoption metrics were part of this engagement, the platform enables genuine cross-institution collaboration while preserving the strict separation institutional data requires.

Closing

This engagement shows Dayo-Tech operating as a strategic partner from a project's earliest stage, authoring the specification, not just fulfilling it, and solving a genuinely hard architectural problem along the way. Enabling collaboration while guaranteeing separation is exactly the kind of challenge that rewards principled infrastructure design, and it reflects how Dayo-Tech builds around institutional trust as much as technical performance.