re:Invent 2015: AWS Is Unstoppable
AWS re:Invent 2015 showcased an ecosystem in overdrive, with new services, aggressive pricing, and a vision for cloud that keeps widening
AWS re:Invent happened last month in Las Vegas, and even watching from my apartment through blog posts, live streams, and Twitter, the scale of announcements was overwhelming. This was the fourth re:Invent, and each year it gets bigger. Over 19,000 attendees this year. Hundreds of sessions. And a torrent of new service announcements that reshape what AWS can do.
Let me try to make sense of what matters.
The Announcement Firehose
AWS announced a staggering number of new services and features. I am going to focus on the ones most relevant to what I study and work with.
Amazon Aurora got significant updates. Aurora is AWS's MySQL-compatible relational database that promises the performance of commercial databases at a fraction of the cost. They announced read replicas across regions, which is meaningful for globally distributed applications that need low-latency reads close to their users. The underlying architecture, which separates compute from storage and replicates data six ways across three availability zones, is a fascinating piece of distributed systems engineering.
AWS IoT was announced as a managed service for Internet of Things devices. It provides device authentication, message brokering via MQTT, and rules engines that can route device data to other AWS services. Given the explosion of connected devices (and the Apple Watch conversation from earlier this year fits here), having managed infrastructure for IoT workloads is increasingly important. The scale challenge is real: billions of devices sending small messages at high frequency require a fundamentally different architecture than traditional web applications.
Amazon QuickSight entered the business intelligence space, offering cloud-native data visualization and analysis. This is AWS moving up the stack, from infrastructure into analytics tooling that business users interact with directly. It signals a strategy of owning more of the value chain, not just the compute and storage layers but the tools people use to derive insights from their data.
AWS Lambda got new triggers and integrations. Lambda, the serverless compute service that launched last year, can now be triggered by API Gateway, SNS, CloudWatch Events, and several other sources. The serverless model, where you write functions and AWS handles all the infrastructure, aligns with a broader trend toward finer-grained compute units. VMs gave way to containers; containers might give way to functions.
Amazon Kinesis Firehose simplifies streaming data ingestion into S3, Redshift, and Elasticsearch. The streaming data pipeline story on AWS is getting more coherent: Kinesis Streams for real-time processing, Kinesis Firehose for loading into data stores, and Kinesis Analytics (announced but not yet available) for running SQL queries on streaming data.
The Ecosystem Effect
What strikes me most about re:Invent is not any individual announcement but the cumulative effect. AWS now offers over fifty services spanning compute, storage, databases, networking, analytics, machine learning, IoT, developer tools, and management. The breadth is extraordinary.
This breadth creates a gravitational pull. Once you are using EC2, it is natural to add RDS for your database and S3 for your static files (as I did with my WordPress POC). Then you add CloudWatch for monitoring, CloudFormation for infrastructure as code, IAM for access control. Each additional service deepens your investment in the ecosystem and increases the cost of moving to a competitor.
From a research perspective, this lock-in dynamic is interesting. Cloud computing was supposed to be about flexibility and portability. In practice, the major cloud providers are building increasingly differentiated services that tie customers to their platforms. The trade-off between convenience (using managed services that handle operational complexity) and portability (being able to move between providers) is a real tension that the industry has not resolved.
Pricing as Strategy
AWS's approach to pricing deserves its own discussion. They have cut prices dozens of times since launch, often pre-emptively, before competitors force them to. The message is clear: we will compete on price, and our scale allows us to sustain margins that smaller players cannot match.
The free tier continues to be one of the most effective developer acquisition strategies in tech. I run my WordPress POC almost entirely on free tier services. When I graduate and start working, I will default to AWS because I already know it. Multiply that by hundreds of thousands of students and early-career developers, and you have a pipeline of future enterprise customers who grew up on AWS.
The reserved instance and spot instance pricing models are also worth noting. Reserved instances offer significant discounts (up to 75 percent) for committing to one or three years of usage. Spot instances let you bid on unused capacity at steep discounts but with the risk that your instance gets terminated if the spot price exceeds your bid. These pricing models map to different workload patterns and risk tolerances, and optimizing across them is itself a research problem.
What re:Invent Tells Us About Cloud Maturity
Four years of re:Invent conferences trace the evolution of cloud computing from infrastructure utility to comprehensive platform.
The early years focused on foundational services: compute, storage, networking. The message was "you can run your servers here instead of your own data center." That was the Infrastructure-as-a-Service era.
Recent conferences have shifted upward. The message is increasingly "you do not need servers at all." Lambda embodies this; you write code, and AWS executes it without you thinking about instances, scaling, or operating systems. Managed services like Aurora, DynamoDB, and Elasticsearch handle the operational burden of running databases and search engines.
The direction is clear: AWS wants to eliminate as much undifferentiated operational work as possible, so customers can focus on their unique business logic. Every piece of infrastructure that AWS manages is a piece of infrastructure that the customer does not need to hire someone to manage.
This aligns with research on cloud economics. The fundamental value proposition of cloud is converting capital expenditure (buying servers) to operational expenditure (paying for usage) and converting fixed costs (maintaining infrastructure) to variable costs (scaling with demand). Each new managed service extends this conversion further up the stack.
Implications for My Research
Several re:Invent announcements connect to my research interests.
The Lambda model raises interesting questions about resource scheduling. In a serverless environment, the cloud provider makes all placement decisions. The customer does not know or care which physical machine runs their function. This shifts the optimization problem entirely to the provider side. How do you schedule millions of short-lived function invocations across a fleet of servers while maintaining low latency, high utilization, and strong isolation between tenants? The research literature on this is still thin, and I think there are opportunities for meaningful contributions.
Aurora's architecture, with its separation of compute and storage, challenges traditional assumptions about data locality. Many scheduling algorithms assume that moving computation to where the data lives is cheaper than moving data to where the computation runs. When storage is a shared, networked service (as it is in Aurora), this assumption breaks down. The optimization landscape changes.
The IoT service introduces edge computing considerations. When devices are distributed globally and generate data continuously, you need processing at the edge, not just in centralized data centers. This distributed processing model introduces new scheduling and placement challenges that differ from traditional cloud workloads.
The Bigger Picture
Watching re:Invent from an academic perspective is a study in how quickly industry moves. Research papers have publication cycles measured in months to years. AWS ships new services on a weekly basis. By the time a paper analyzing a particular AWS service goes through peer review, the service might have changed significantly.
This speed differential is both a challenge and an opportunity. It is a challenge because academic research risks being perpetually behind the state of practice. It is an opportunity because the rapid pace of innovation creates a need for rigorous analysis, theoretical frameworks, and systematic evaluation that industry, in its rush to ship, often neglects.
I came back from watching re:Invent (virtually) with a renewed appreciation for the pace of cloud innovation and a clearer sense of where research can add value. The fundamental problems of resource management, scheduling, and optimization do not go away as the abstraction layers change; they just manifest differently. That persistence is what makes this research area both durable and endlessly interesting.