Machine learning, storage, compute and shot at Microsoft dominate new announcements at AWS re:Invent

AWS Re:Invent may have been virtual this year, but two things didn't change: a three hour keynote unveiling new services from CEO Andy Jassy, and dozens of new services.

While this year’s AWS re:Invent event was virtual, it still went according to form. CEO Andy Jassy still gave his normal three hour keynote, and announced many new services – over twenty in total.

Jassy led into his introduction of new and enhanced AWS services by emphasizing his customary theme of reinvention. He noted that only 17% of 1970s Fortune 500 companies remain on the list, and only 50% from 2000.

“Companies often have to reinvent themselves several times,” he said. “Companies on the verge of falling apart often try to reinvent themselves. It’s a crapshoot if you wait that long. You want to be reinventing when you are healthy.”

Jassy then turned to AWS announcements around compute.

“Compute is continuing to be reinvented as we speak,” he said. He noted that this involves three forms: instances, containers and event-driven serverless computing.

“These three modes of compute are here to stay,” Jassy said.

He gave a shoutout to the announcement from Monday night’s late night keynote, the announcement of new Mac instances for Amazon EC2 that will let Apple developers natively run macOS within AWS.

“This will let Mac developers access the cloud much more easily,” Jassy stated.

After recounting AWS trumps with the introduction of AWS Inferentia’s inference-focused machine learning, which now handles 80% of Alexa’s predictions, Jassy stated that AWS had not forgotten about machine learning training, and made two new announcements.

“We will offer next year, in the first half of the year, Habana Gaudi-based EC2 instances in partnership with Intel,” he said. “These will provide 40% better price performance than the best GPU EC2 machine learning training instances.”

The other new offering here is AWS Trainium, which will be available in the second half of 2021.

“This is our machine learning training chip to deliver the most cost-effective training in the cloud,” Jassy noted.

Jassy then turned to the smaller units of compute – containers and serverless – that are growing in popularity.

“For containers we have EKS, for Kubernetes at scale, ECS and Fargate, our serverless container offering, which no one else has anything like,” he said. ”All three of these are growing like weeds, not just EKS. Many customers start with Fargate because it’s so easy to get going. Different teams have different preferences and different use cases.”

Jassy said that people have asked for these same management capabilities on-prem. Accordingly, he announced Amazon ECS Anywhere and Amazon EKS Anywhere.

“Amazon ECS Anywhere lets you run ECS in your own data centre,” he stated. “It lets you have the same ECS APIs and cluster management on-prem, and works on any on-prem infrastructure.” Amazon EKS Anywhere does exactly the same for Kubernetes.

“These are both coming in 2021, but to get you started now, we are open sourcing our EKS distribution to you so you can transition for EKS Anywhere now,” Jassy stated.

“We think our container customers will be excited by these, but more and more customers are using event-driven serverless computing, which we pioneered with Lambda,” he said. “We are now launching Lambda Container Support, which lets you build Lambda-based applications using existing container development workloads. It changes the ability to deploy Lambda.”

Next on the list was AWS Proton, a new service for building microservices.

“We have been asked to manage the deployment of smaller instances of compute,” Jassy said. “There hasn’t been anything out there that helps customers manage this deployment challenge. AWS Proton is the first fully managed service for microservices, with everything needed to deploy a microservice except the actual application code. This is a gamechanger with regard to deploying containers and serverless apps.”

Jassy then turned to the topic of the reinvention of data and data stores.

“In every hour, we create more data today than in an entire year 20 years ago,” he indicated. “Old tools and data stores aren’t going to cut it.”

Jassy noted that block storage is foundational and pervasive, is faster because of no metadata, and is used in most EC2 use cases. The current version of Amazon Elastic Block Store volumes, EBS Gp2, was built in 2014.

“Feedback in the last year or two has been that you would like the cost to be less, and would like to scale IOPS without having to scale storage with it,” Jassy said. He then announced Amazon EBS Gp3 volumes, the next generation of the offering, that provides the ability to provision additional IOPS without adding storage, and is priced 20% lower per GB.

Next came Amazon EBS io2 Block Express volumes, a SAN in the cloud for high performance needs.

“Io2 is for even higher IOPS and throughput, like Oracle and SAP HANA,” Jassy said. “Customers have been striping together Gp2 to handle this, but the more you do that, the harder it is to manage. So customers were forced to use SANs on prem. There has been no SAN in the cloud – until now.”

Io2 Block Express is the first SAN built for the cloud, with 4x the IOPS, throughput and storage capacity of Io2.

“It is SAN performance without the headaches,” Jassy said. While the SAN features will be initially limited, more will be coming in 2021, including multi-attach, I/O Fencing, and making elastic volumes work with it.

“This is a huge game changer for your most demanding applications,” he stated.

Jassy then turned to databases, and in particular making it easier to move from on-prem databases – especially SQL – to AWS Aurora.

“On-prem requires a lot of heavy lifting, but the overwhelming majority of relational databases live on-prem in Oracle and SQL,” he commented. “It’s an unhappy place. It’s expensive and proprietary and they can change licensing midstream. Microsoft put a tax on moving from their on-prem SQL to AWS or Google Cloud.”

Jassy said that application code being tied to proprietary databases makes it harder to move workloads to Aurora, especially with Microsoft cracking down against moving anywhere other than Azure.

“So we are announcing the launch of Babelfish for Amazon Aurora PostgreSQL,” Jassy said. “It lets you run SQL servers, and understands Microsoft’s proprietary schemes, so you can stop paying for SQL licenses.

“We also realized this was bigger than Aurora Postgre, so we decided we would opensource Babelfish for Postgres,” Jassy added. This will use the Apache 2.0 license in GitHub.

“It will be a huge enabler for customers to move away from legacy databases,” Jassy stated.

Jassy also noted that while Amazon Aurora has been the fastest growing AWS service ever since its launch, with over 100,000 customers, Aurora customers asked AWS to run Aurora in a serverless architecture, and the result was Aurora Serverless.

“Now we are announcing the launch of Aurora Serverless V2.0, which allows scaling to hundreds of thousands of transactions in a second,” Jassy said. “It totally changes the game for you because it now lets you scale up instantaneously. The database companies won’t build something like this because it would cannibalize their core offerings.”

Jassy also introduced AWS Glue Elastic Views to respond to customers wanting more freedom for data to move around.

“This lets you build materialized views that automatically combine and replicate data from different data stores,” Jassy said. “It allows you to set up a materialized view and move it from a source data store to a target data store – and if the data changes in the source, it will change it in the target store as well. This is a huge gamechanger in being able to move data – which was so much work in the past that people rarely did it.”

Next came machine learning, with Jassy noted is hand in hand with what’s happening with data. He told the virtual audience that he won’t speak for 75 minutes on machine learning, as he did last year, because this year there is a separate keynote on that. There were, however, quite a few machine learning announcements.

“We were asked to make doing data preparation for machine learning easier,” Jassy said. “It’s hard, and converting features takes a lot of time. There’s a lot of writing queries and code. So we are announcing the launch of Amazon SageMaker Data Wrangler, the fastest way to prepare data for machine learning. You point it at the appropriate AWS or third party data store, and it identifies the right transformation to make, which you can preview easily in SageMaker Studio. It manages all the work under the covers. This is a total gamechanger in the time it takes to do data prep for machine learning.”

Jassy also unveiled SageMaker Feature Store, which makes it easy to store and share Features with teams, and SageMaker Pipeline, the first purpose-built CI/CD service for machine learning

“You can quickly create machine learning workloads and automate a number of the steps, and SageMaker Pipeline manages all these dependencies between the steps,” Jassy noted.

For machine learning for those who don’t want to build models, Jassy introduced DevOps Guru.

“This is a new machine learning service that automatically detects operational issues and recommends actions,” he said. In cases like missing or misconfigured alarms, resources approach limits, overutilization of databases, or memory leaks, we will notify when we see a problem, and recommend remediation.”

For customers asking not to have to put the pieces together, Jassy introduced Amazon QuickSight, which he said finally gets natural language query right.

“Natural language query hasn’t had good results,” he said. “Amazon QuickSight lets you ask any question in natural language and get an answer in seconds. You type a question in the search bar, and it uses Natural Language Processing to understand domain specific business language.”

Multiple new capabilities were also announced for Amazon Connect, their omnichannel cloud contact centre.

“Real Time Contact Lens uses machine learning to detect customer experience issues during live time,” Jassy said, indicating that this will let supervisors help out the agent or even take over the call. “It was not easy to do, which is why it took our team a year to do it.”

Amazon Connect Tasks automates and tracks tasks to make follow-up tasks easier for agents and lets some tasks be automated entirely. Amazon Connect Voice ID provides real time call authentication without disruption. It requires the customer opting in ahead of time, and uses machine learning to build a voice print for each customer. Amazon Connect Wisdom gives contact centre agents more real-time information, while Connect Customer Profiles provides agents with a unified profile of each customer, with Jassy pointing out that it can he helpful to know if a customer calling about a complaint is the same person behind a major pilot involving the call centre.

Jassy then moved to new offerings targeted at more industrial machine learning uses.

“Manufacturing and industrial companies tell us they can change customer experience using machine learning, but they don’t have the equipment and talent to make that happen,” he said.

Amazon Monitron is a new end-to-end solution for equipment monitoring to enable predictive maintenance.

“We use this to show you where there are anomalies to indicate where you need to do predictive maintenance,” Jassy noted.

Amazon Lookout for Equipment provides anomaly detection for industrial machinery.

“We assess sound vibration and temperature, build a model of normal anomalies, and send alerts for deviations,” Jassy indicated

AWS Panorama Appliance is an appliance that lets computer vision be added to existing on-prem cameras.

“It picks up concurrent streams for cameras and identifies video streams, to add computer vision to existing onsite cameras,” Jassy said.

AWS Panorama SDK lets hardware vendors build cameras that will be better smart cameras moving forward.

Two new formats for Amazon Outposts were also announced, both small sizes compared to the original rack Outpost. One is a 2U version, and the other a 1U that is the size of a pizza box.

Finally – Jassy announced 12 new AWS Local zones, all of them in the U.S, which will roll out in 2021.