re:Invent - Recap

5. dec.

re:Invent - (the biggest AWS conference of the year) Recap: A Glimpse into the Future of Cloud Computing with AWS

Every year in Las Vegas AWS holds its annual re:Invent conference, an opportunity for AWS enthusiasts, customers, engineers, and partners from all over the world to come together, and inevitably, for AWS to announce their latest innovations and services.

This year has been no exception. AWS re:Invent, spanning from November 27th to December 1st, showcased a prominent emphasis on advancements in both serverless capabilities and, unsurprisingly, the introduction of new AI and machine learning services, reflecting the evolving landscape of the past 12 months. Here is a summary of new service announcements that are worth knowing about.

Amazon Bedrock. As an organization, imagine harnessing the most advanced foundation models available in the market for your production-level systems, enabling you to avoid the hassle of constructing your own model and instead benefit from the expertise of industry leaders. With Amazon Bedrock you can do just that; customize industry leading FMs from AI21 Labs, Amazon, Anthropic, Cohere, Meta, and Stability AI using your organization’s own data. This service was made generally available in September 2023, and by the end of November 2023, AWS already had over 10,000 customers using Bedrock. Bedrock is democratizing machine learning models to everyone - you can get going very quickly, and no prior knowledge of machine learning is required.

Amazon Bedrock Agents. An agent helps an end-user complete actions based on organizational data and user input. Agents orchestrate interactions between selected AI foundation models. Developers can integrate these agents to accelerate the delivery of generative AI applications, saving weeks of development effort. This service is now generally available.

Guardrails for Amazon Bedrock. Allows you to effortlessly set up harmful content filtering in alignment with your organization's responsible AI policies. This service addresses prevailing concerns about responsible AI in the marketplace, marking a positive stride forward. For instance, an e-commerce company can now safeguard against the use of hate speech by a chatbot, and a utility company can prevent the inadvertent disclosure of personally identifiable information (PII) during a customer service call. This service is available in preview today.

Amazon Q. A generative AI-powered assistant for your work-place has been announced and is in preview. This AI-powered work assistant can be tailored to your business, so the answers it generates encompass the context of both your organization as well as a specific role in that organization. It’s also an expert in AWS well architected patterns, so you can ask technical cloud infrastructure questions in the context of your business. Let's face it—given the vast array of over 200 AWS services available to date, it's practically impossible for anyone, except perhaps a robot, to claim absolute expertise in all of AWS's offerings, let alone the potential integrations between them all. Well, this robot has arrived, and the robot’s name is Q. We can leverage Q to help make informed architecture decisions. Additionally, if you are troubleshooting errors in the AWS console, you will now have the option to press a button ‘Troubleshoot with Amazon Q’ for error resolution suggestions. Q can also be integrated into a developers IDE where it can be used to query the code base, as well as used for generating code and unit tests. I’ve started using Q in my IDE (VS Code), and I recommend trying it out. Q comes shipped with the AWS Toolkit extension for VS Code, and it’s free, so download the toolkit and get started. Q has also announced a feature development capability for developers that is now available in preview in AWS Code Catalyst, and will soon be available in the IDE. Watch this space!

AWS Aurora Limitless for PostgreSQL and MySQL. A truly serverless relational database service that boasts serverless sharding for automated horizontal scaling of your database. They’ve accomplished this by building high precision time synchronization into their Nitro chips across all hardware that removes clock skew, which enables AWS to keep an accurate database transaction log using timestamps, as well as leveraging a new type of hypervisor they’ve dubbed Caspian, and a feature called the Caspian heat management system that enables ‘cooperative oversubscription’, which is the ability to scale database instances up limitlessly on shared tenancy. This service is currently in limited preview, and available with PostgreSQL, and eventually MySQL as well.

Amazon Elasticache Serverless. You heard it. Amazon Elasticache now has a serverless option. Thanks to the infrastructural innovations mentioned above for Aurora Limitless, these improvements have been leveraged to benefit other AWS services, including Amazon Elasticache. This means no more hassle with underprovisioning or overprovisioning cache servers yourself. AWS handles the hassle of it all for you, and in true serverless fashion, you only pay for what you use. This service is already generally available for AWS customers today.

Data warehouses. Amazon Redshift Serverless was released in 2021, so you can run data analytics at any scale without having to manage data warehouse infrastructure. It’s now been improved using next generation AI-driven scaling and optimization. Essentially, this revamped service does a much better job proactively anticipating and scaling the data warehouse based on predicted capacity, so your BI dashboards aren't slowed down by anomalously large queries made by the data science team using the same data warehouse, for example. The machine learning model uses over 50 query feature embeddings to predict if a query is anomalously large and therefore should have its own additional infrastructure allocated to it, thereby scaling out your Redshift cluster automatically for you, and scaling back afterwards.

S3 Express One Zone. A new S3 (blob object storage) tier to provide the lowest latency available for your most frequently accessed objects is now available. For the first time you can select an availability zone (AZ) for S3 storage (hence ‘One Zone’), which enables you to move your most frequently accessed objects in storage into the same AZ as your high performance compute resources. This storage tier is 10x faster than object retrieval in the S3 standard storage tier. Not only this, S3 Express Once Zone results in reduced storage costs by at least 50% compared to S3 Standard storage tier. Make sure your disaster recovery plan accounts for this, though. All your data is stored in one AZ at this point, and you’ll probably want this replicated to another AZ for your RPO (recovery point objective).

AWS Graviton 4. AWS has announced a fourth generation for their own server processor that is faster and more energy efficient than ever before. To make use of these memory optimized instances, they are available in preview as an R8g EC2 type. These are well suited to large database workloads or large data analytics workloads.

Amazon EC2 Capacity Blocks for ML. As an industry first, you can now reserve UltraClusters (with hundreds of GPUs) in short bursts (1-14 days) to process your machine learning workloads. This is ideal for fine tuning foundation models (FMs), or for short duration workloads, or for handling capacity surges.

AWS Trainium 2. While other cloud providers are still talking about producing their own tailor made chips for generative AI workloads, AWS released the second generation of their Trainuim chip. They are 4x faster than the first generation, and provide 65 exaflops of on-demand supercomputing performance. Wondering how much an exaflop is? Well a typical laptop can process about 100 gigaflops (10^9 floating point operations per second). An exaflop is 10^18 floating point operations per second. So 65 exaflops is the equivalent of about 42 million laptops worth of processing power per second.

AWS Neuron. AWS’s SDK to optimize machine learning on Trainium now boasts support for 93 out of the world’s top 100 models. Neuron supports your favorite machine learning frameworks such as PyTorch and Tensorflow (and soon, JAX), so you can get running on your machine learning workloads with only a few lines of code.

Zero ETL integrations. What exactly is zero ETL? AWS has a vision for an ETL free future, where AWS users perform analysis on their data without having to handle the extract, transform, load (ETL) from their data sources. Amazon announces three more 'Zero ETL’ integrations for Redshift, and one for OpenSearch. Zero ETL Redshift integrations are now available for Aurora PostgreSQL, RDS for MySQL and Amazon DynamoDB. With a few clicks you can now perform large scale analysis directly in Amazon Redshift from the aforementioned data sources with zero ETL requirements. A zero ETL integration between DynamoDB and Amazon OpenSearch service was also announced. These zero ETL integrations are generally available today.

Software vulnerability scanning of containers in your CI/CD pipelines. Amazon Inspector has introduced enhanced security capabilities, and you can now scan your containers at build time directly in your deployment pipelines. Very nice!

That sums it up neatly. The prevailing theme for the newly announced services at this year's re:Invent revolves around advancing the abstraction of configuration and infrastructure for customers. This accomplishment results from a synergy of innovative hardware and cutting-edge AI and machine learning technologies. AWS serverless is positioned to become even more intelligent as machine learning models evolve and AI is increasingly harnessed. Additionally, this year's re:Invent unsurprisingly marked a significant step forward in making machine learning services more accessible and democratized. At C Nation, we eagerly anticipate the transformative impact of these developments on our customers' businesses.

AWS stands out as the undisputed global leader in cloud services, with three times as many data centers and 60% more services than the next largest cloud provider. This advantage not only ensures higher resilience and availability for our customers but also empowers them to innovate at the forefront of cloud technologies. Partnering with the world leader in cloud computing, AWS, provides unparalleled opportunities for success. Don't hesitate to contact us with any questions; we'd be delighted to assist you on your cloud journey, especially with the latest advancements and increased accessibility in the fields of machine learning and AI.

For further information on the announcements at this year’s re:Invent see the following:

Amazon Bedrock: https://aws.amazon.com/bedrock/

Amazon Bedrock agents: https://docs.aws.amazon.com/bedrock/latest/userguide/agents.html

Guardrails for Amazon Bedrock: https://aws.amazon.com/bedrock/guardrails/

Amazon Q: https://aws.amazon.com/q/

Amazon Q feature development capability: https://aws.amazon.com/about-aws/whats-new/2023/11/feature-development-capability-amazon-q-preview-codecatalyst/

Amazon Aurora Limitless: https://aws.amazon.com/about-aws/whats-new/2023/11/amazon-aurora-limitless-database/

Amazon Elasticache Serverless: https://aws.amazon.com/blogs/aws/amazon-elasticache-serverless-for-redis-and-memcached-now-generally-available/

Amazon Redshift Serverless: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-serverless.html

S3 One Zone: https://aws.amazon.com/blogs/aws/new-amazon-s3-express-one-zone-high-performance-storage-class/

Memory optimized EC2 instances (Graviton 4): https://aws.amazon.com/blogs/aws/join-the-preview-for-new-memory-optimized-aws-graviton4-powered-amazon-ec2-instances-r8g/

Amazon EC2 Capacity Blocks for ML: https://aws.amazon.com/ec2/capacityblocks/