Towards an infinite laptop

The new Anyscale platform offers the ease of development on a laptop combined with the power of the cloud.

During a series of short keynotes at the Ray Summit this morning, Anyscale1, the company formed by the creators of Ray, publicly shared their initial product offering. Dubbed the “infinite laptop”,  Anyscale’s platform allows developers to treat their laptop as an “infinite cluster”. Developers can use their preferred development tools (e.g., IDE, notebook, text editor, etc.) on Anyscale’s platform, and seamlessly burst into a cloud platform when needed. 

The demo at the conference walked the audience through the process of a building an end-to-end application: training and deploying models for a recommendation engine. Consider a developer in the process of building and training a model for a machine learning application. The challenge with machine learning is that the model building phase requires multiple runs. You usually need to explore different models and model parameters before settling on a model. If model training takes too long, it limits the amount of ideas a developer can try. On the other hand, developers invest time setting up their laptops and they feel productive when their favorite IDE, libraries and software are in place. Ideally one can explore ideas on a laptop (say a simpler model, or a model that uses less training data) and burst out to a compute cluster when computations begin to overwhelm your laptop. The Anyscale platform delivers this exact solution!

The Anyscale platform gives developers the “development experience on their laptop combined with the power of the cloud”. This means that Anyscale syncs code on your laptop to the cloud. You can continue to edit your program using the development tools on your laptop and Anyscale will optionally execute your code on the cloud. Machine learning developers can use specialized processors (including TPUs or fast GPUs) and scale out to a cluster if they need to parallelize certain computations.  It is cloud native and you can even use multiple clouds at the same time. You really have the best of both worlds: the tools and libraries that you feel most comfortable with, backed with the compute resources you need to be productive and effective. My immediate reaction after watching the demo was “That’s exactly how I want to work and the platform I need!”

Image: Edward Oakes, Anyscale Product Demo at the Ray Summit

Building efficient and performant distributed applications – the sort needed in modern AI – is challenging for most developers. A developer needs to be able to deal with failures (which are inevitable), control where computations take place, and efficiently handle multiple processes that handle data and communicate with each other. Fortunately, developers need not start from scratch. As Anyscale co-founder Robert Nishihara highlighted in his keynote, the open source project Ray is becoming the software platform for building distributed applications.

Image: Robert Nishihara at the Ray Summit (“The Future of Ray”)

Anyscale provides what co-founder Ion Stoica described as a “serverless experience without serverless limitations”. It abstracts away servers and DevOps so developers can focus on writing their applications.  And since Anyscale uses Ray, the Actor model allows it to support stateful applications that are as fast as those built from scratch by experts in distributed programming.

Image: Ion Stoica at the Ray Summit (“Programming the Cloud as Easily as your Laptop”)

Why is this exciting? Machine learning and AI are becoming central to many applications. But building machine learning applications can be extremely compute intensive. Even with the emergence of hardware accelerators, many machine learning workloads go beyond the capacity of a single server. We need tools that enable regular developers to quickly build distributed applications. The Anyscale platform is a great step in this direction.

The company announced this morning that the Anyscale platform is currently available as an invite-only Beta. Learn more by watching the following demo from the Ray Summit. [Update 2020-10-08: The official announcement blog post from Anyscale is now up.]

If you’re interested in Ray, read this new post describing the version 1.0 release.

[1] Full disclosure: the author is an advisor to Anyscale.

Related Content: Download your free copy of the 2020 NLP Industry Survey Results.

%d bloggers like this: