Start in minutes with $100 of free compute
What would you build with a virtual datacentre?
The Distributed Compute Protocol is a lightweight, powerful framework for HTC applications. It is a cross-platform standard for computing that makes the same code run on all hardware.
DCP lets users provision and share heterogeneous compute resources across any network, from the office to the globe. Even desktops, laptops, and smartphones can seamlessly power virtual clusters.
While there are many frameworks for distributed and grid computing, DCP brings this into the 21st century.
Previous solutions were not standardized for different hardware and networks, while DCP executes identical code anywhere. It gives developers a way to build arbitrary applications rather than confining them, and makes these shareable with the entire world. Finally, DCP combines both on-premises networks with the cloud to put more computing power in reach than ever before.
DCP’s Compute API is the easiest way to express arbitrary parallel workloads. It is used everyday by the same developers who built it for AI, Computational Finance, Bioinformatics, and more.
Compute and network resources are completely abstracted from the underlying hardware, so you always get seamless performance even across different clouds.
DCP uses ultra-lightweight web technology that gets closer to bare metal performance than any other. A workload run on DCP is commonly more efficient than the same run in a Container or VM.
DCP allows you to connect and disconnect your own hardware to build high performance compute grids. Additionally, compute grids from datacentres and other institutions are publicly available through the Distributed Computer. These two models are known as Compute Group and the Distributed Computer.
Show Me
Accelerate high-throughput computing jobs with underutilized hardware you already own. Any CPU or GPU can be connected to build a fully functioning compute grid across a single LAN or even an international WAN.
The ultimate private cloud powered by a global federation of compute providers, the Distributed Computer finishes the toughest jobs in minutes. Access thousands of cores with a second’s notice, all without managing a single Container or VM.
Be one of the first to build cutting edge applications with free compute on DCP. The First Devs Program will send you all the latest updates, and give you privileged first access rights when you join today!
From deep neural networks to multiphysics simulations, DCP powers cutting-edge applications.
Access a larger pool of compute resources both on-premises and from the cloud without managing a single server.
Code becomes fully interoperable and portable with the first framework that runs everywhere, from single board computers and the IoT to enterprise servers.
DCP is lighter-weight and far more responsive than Containers or VMs, and requires almost no programming experience to master.
DCP lets nodes easily join and move between software defined clusters. Its overlay network is always working hard to optimize your jobs.
Harness the advantages of a distributed system with uninterrupted service and security. DCP delivers results in real-time and with an integrated failover system.
Customize your workflow with DCP 3rd party apps, or build your own. Reach a global audience with your software and earn royalties for every use.
Build and deploy your application from DCP's browser-based IDE. Building powerful compute clusters is as easy as configuring a few settings in a web browser.
Effortlessly build applications to take advantage of thousands of cores. The compute.for( ) command has unmatched support for data parallelism at any scale.
DCP is useful for fully integrated workflows, offloading compute-intensive steps, or handling requests from other apps. Every application bounded by HTC will benefit.
Get US$100 of free compute for your first workload on DCP.
See All AdvantagesDCP is a complete distributed OS built to be standards-based, high performing, and future-proof.
The Compute API uses a memory safe generator called compute.for( ) to express parallel workloads. A developer only needs to identify the ideal unit size of their data to process in parallel, and write the methods to execute.
Each distinct parallel unit of work in DCP is called a 'slice', which contains both the data and code that is to be run. On a CPU, one slice generally maps to one core. These slices are then distributed to the proper hardware to be computed.
DCP uses a lightweight JavaScript engine. It is more lightweight than Containers or VMs, and completely removes the complexity of managing infrastructure.
An instance of DCP on one machine is called a Worker. The Worker is a Node.js daemon that coordinates the developer’s workload and transmits the results back. Starting a new Worker is as easy as opening a web page, and has these benefits compared to Containers and VMs:
> Lower memory overhead.
> Faster response times, including for cold-starts and application switching.
> Greater flexibility, being able to run across any fleet of hardware.
> Improved security from having no I/O capability.
DCP applies its own algorithms to solve the old problem of scheduling jobs on a distributed network.
The DCP Scheduler is the primary module that does this, broadcasting both time and space-sharing commands while avoiding common issues like thrashing.
The Scheduler takes multiple factors into account such as speed, energy use, and even the level of CPU and memory capacity it should not exceed. Because DCP is suitable for all environments, the Scheduler works across everything from bare metal to Kubernetes deployments.
See how DCP is an accelerator for parallel and distributed computing.
Show Me The TechThe Distributed Computer is a global cloud federation powered by everything from universities to enterprise datacentres. It is as easy to access any of these resources through the Compute API as programming your laptop.
With the Marketplace, your code can:
Power applications run by thousands of paying users
Load balance your apps during intensive operations
Automatically scale across thousands of cores on-demand
Leverage compute accelerators like GPUs & NPUs without needing a rewrite
DiskFit is a web app made with DCP that helps astrophysicists understand how galaxies form. Originally in Fortran, it now runs in Node.js with DCP.
Learn how what used to take 3.5 days on a single PC now takes minutes, and how DCP can do the same for your application.
DCP is always improving with help from its community and core development team. The project has come a long ways, but there are many exciting developments yet to come. Stay tuned for stateful computing, support for popular integrations, and coarse-grained parallelism!