Approach

A high-level view of how Vishwanath Grid is put together. Operational documentation lives in the operator handbook.

Federated compute, one interface

We pool compute capacity across multiple partner sites. Each site keeps physical control of its own hardware; the grid orchestrates job placement across the federated pool so a researcher submitting from MIT Pune can run on a node at any partner campus and never know the difference.

flowchart LR
  R[Researchers] -->|single sign-on| GW[Vishwanath Grid edge]
  GW --> S[Scheduler]
  S --> P1[Partner site 1<br/>compute nodes]
  S --> P2[Partner site 2<br/>compute nodes]
  S --> P3[Partner site 3<br/>compute nodes]
  GW --> D[Shared storage]
  GW --> C[Collaboration tools]

A single login, every tool

Researchers do not learn ten different passwords. A single identity carries them from the job scheduler to the data store to the project management tools to the dashboards. Sign in once; the cookie carries you through everything.

Open-source, top to bottom

Every component of the stack is released under a permissive license (AGPL, Apache 2, or MIT). The choice is deliberate: when a grant ends, when a vendor pivots, when an organisation forks — the platform keeps working. Nothing in our stack can be held hostage by a renewal cycle.

Designed for the long horizon

Universities run on five- and ten-year horizons; commercial cloud billing does not. Vishwanath Grid is built for the multi-decade research-compute timescale: hardware that can be re-tasked, software that does not require a subscription, and a federation model that survives any individual partner leaving.