Spotinst adds automation to simplify and optimize Kubernetes clusters


As a leading purveyor of excess hyperscaler capacity, Spotinst has years of experience in algorithmically maintaining cost-oriented clusters for production workloads. In December 2018, it launched Ocean, a serverless container engine to help customers get started with Kubernetes, using sensible defaults and enabling ‘nodeless’ operation with autoscaling at the pod level. The company’s pay-as-you-save pricing is a differentiator: In a market where bigger deployments and more cloud spending typically yield higher revenue for service providers, Spotinst brings in more when customers spend less.

Spotinst stay one step ahead of enterprise demand, anticipating which technologies will take hold and benefit from automation while creating backstops to smooth the onboarding of hardto-refactor workloads. It started in 2015 with spot instances – VMs drawn from excess hyperscaler capacity at steep discounts (up to 90%) versus on-demand rates. While tuning its algorithms to safely maximize the use of this inexpensive capacity for production web workloads, the company moved on to containers, solving for the complexity of Kubernetes with an abstraction layer that automates cluster scaling in keeping with application-level requirements.

Next on the roadmap are big-data and data science workloads, which Spotinst expects to become integral to day-to-day development in the coming years. The goal is to give enterprises a single interface for managing cloud applications with dynamic provisioning and orchestration – all while optimizing the infrastructure layer for cost and performance.



In the four years since its inception, Spotinst has honed its ability to use low-cost ephemeral instances – called spot instances on AWS, preemptible VMs on Google Cloud Platform, low-priority VMs on Azure, spot instances on Alibaba Cloud and transient servers on IBM Cloud. Spot instances are often priced at discounts of 80% or more versus on-demand rates, but they come with a catch –when the provider needs to reclaim the capacity, it can terminate the instance, in some cases without warning. Spotinst uses predictive analytics to forecast when excess capacity is likely to be available, then dynamically pools spot resources with on-demand and reserved instances (RIs) to create resilient, cost-effective compute clusters that are suitable for production environments, which can yield savings of up to 80% versus on-demand rates. Few enterprises have the stomach to manage spot pricing manually or with the help of provider tools. 451 Research’s survey found that only 9% of enterprises were using spot resources to pay for public cloud services, versus 52% paying for ondemand usage billed in arrears.


How enterprises pay for public cloud usage


Spotinst positions itself as a cloud workload automation platform, picking up where static provisioning tools and forensic monitoring and cost management software leave off. Its flagship Elastigroup product stands out among cost-optimization tools by using predictive analytics to automatically assemble compute clusters for high availability and low cost. The product is available for AWS, Google Cloud Platform and Microsoft Azure.

At the end of 2018, Spotinst introduced Ocean, a serverless container-as-a-service platform, to intelligently provision and optimize container clusters, using technology similar to Elastigroup to algorithmically run them on a combination of spot, on-demand, and reserved VMs for cost and performance. Ocean expands on Kubernetes’ ability to do autoscaling at the pod level, managing underlying servers and worker nodes in keeping with application-level needs; as with Elastigroup, Ocean is available for AWS, Google Cloud Platform and Azure.

Spotinst has implemented heuristics and AI to enable per-container showback for detailed activity, such as the cost of traversing the network and storage, collecting information to feed into its cost-optimization routines. The company says Ocean has been enthusiastically received and now accounts for 30% of revenue. Spotinst is releasing three new services to give customers end-to-end support as they plan and execute migrations to its platform.

1/ Cloud Analyzer is an assessment tool that discovers resources within an existing environment and finds opportunities for performance and savings improvements.

2/ Spotinst Eco, based on technology acquired with StratCloud, does for reserved instances what Elastigroup does for spot, using machine learning to optimize an organization’s use of reserved capacity and bringing flexibility to RI purchasing.

3/ Managed Instances makes it possible to run single-instance, stateful workloads on low-cost excess capacity, enabling spot savings for lift-andshift, database or dev/test workloads.

By the end of 2019, Spotinst plans to launch a platform (already in beta with some customers) for cost-effectively managing the infrastructure beneath workloads using Spark, TensorFlow, Jupyter and other big-data tools and frameworks.


Business model

 A typical process starts with assessing a customer’s cloud accounts and identifying a single environment for proof-of concept testing of Elastigroup. The software tracks performance, and typically shows immediate savings by using spot instances. Once the customer has realized the likelihood of achieving a return on investment with Spotinst,it is followed by production workloads, integration with customer CI/CD platforms, the migration of more accounts to the platform, and ultimately adoption of more products and a greater variety of workloads.

Spotinst recently introduced a free tier that allows users to manage up to 20 nodes on the platform. The Professional plan charges on a pay-as-you-go basis by collecting 20% of the costs saved versus running the customer environment at on-demand prices; a fully featured 14-day free trial is available.


Source: 451 Research