AWS re:Invent 2022: Perspectives for the CIO | Live Blog

AWS re:Invent 2022: Perspectives for the CIO | Live Blog

I’m here at The Wynn hotel in Las Vegas as the huge AWS re:Invent 2022 confab kicks off today. The event catches the cloud services giant at a historic inflection point, as the cloud hyperscaler industry enters early maturity and the epic capital bets Amazon has made over the last decade and a half are showing their wisdom (or not.) The wildcard? This is happening right as the global economy hits significant speed bumps for just the second time in AWS’s history. Chief Information Officers (CIOs) are closely scrutinizing their cloud portfolio this year and next to ensure they get the most value from them. What they learn at re:Invent this week will certainly have impact on their cloud trajectories.

Through the course of this next week, I’ll be covering the main re:Invent track as live as I can here and Twitter, as well as the separate but co-joined AWS Analyst Summit. I’ll be providing real-time analysis and coverage, mostly from a CIO perspective but also from an enterprise architecture point-of-view. Refresh this blog to pull the latest.

Monday, November 28th, 2022 – 8:15am

First up, I’m heading downstairs at The Wynn for the first session of the AWS Analyst Summit which is a public cloud roundtable. This session is off the record in terms of details but I’ll provide oblique thoughts as it progresses.

8:23am: It’s clear that Amazon is very proud of its Graviton 3 custom silicon. It’s 60% more energy efficient over x86. Custom silicon has become a prime differentiator as the super hungry machine learning and AI workloads businesses want to run consume vast horsepower in order to provide competitive advantage. Accelerating the performance of databases is also a major use case. Large language models that support intelligent conversations is another, and the training requirements with vast models to drive performance and reduce cost of using today’s very large models. Graviton first emerged in 2018, and the latest iteration provides some real price/performance advantages over general purpose CPUs. 

For the CIO, I anticipate that the increasingly convergence of overarching ModelOps strategy with cloud strategy will begin to significantly emphasize putting custom silicon into the center of enterprise-wide ML and AI operations.

8:30am: Networking is also providing to be vital capability, the the hyperscalers are driving cutting-edge advances in cloud networking. Expect to see a number of announcements around this during re:Invent this week.

8:37am: High performance computing (HPC) is also a hot topic this week and is clearly an area where Amazon feels like it can compete at a very high level of competency. It’s also likely where significant margin is as well. However, HPC instances are real investments (they aren’t cheap on an hourly basis) but they do bring the elasticity and scale of cloud to the most difficult compute problems likely to be at the forefront of strategic investments in IT and digital capabilities in Amazon’s customer base.

8:41am: “Over-provisioning” is another phrase coming up a lot here at re:Invent. Downsizing to different, smaller instances or processor types in a seamless and automatic way is going to be a very interesting topic for the CIO and cloud operations teams. AWS’s Trusted Advisor seen as not having enough granularity to really help richer, hybrid or Outpost environments. Lots of interest in cutting costs while also managing sustainability. Moving workloads to the cheapest carbon neutral instance type. 

8:46am: Bringing the entire AWS platform — and any other hybrid cloud resources — into one single virtual private cloud (VPC) is seen as a ‘holy grail’ of contemporary cloud. But there is a concern that AWS tools are emerging more to fix point barriers to achieving this, rather that providing a well-designed, overarching VPC capability with minimum complexity and optimal efficiency/manageability. Orgs continue to be hungry for capability multicloud/cross cloud tools to manage their cloud estates. Developers are now being forced to understand cloud networking, cybersecurity, and provisioning to a far greater level of detail than they ever wanted to. I expect announcements at both the tactical and strategic level for this during re:Invent this week (I don’t have any specific knowledge of this.)

8:51am: Running very large workloads is another issue at the forefront of hyperscaler services. Instances sizes within AWS are getting both smaller as well as much larger, in order to fit large models that have hundreds of millions or even billions of parameters in a single. Local zones are continuing to see expansion to bring instances closer to large populations. At the same time, there are questions about whether AWS is going to consolidate its sprawl of zones and services, but it a very challenging topic, even as many customers would like to see more simplicity in AWS services design and consumption.

9:00am: Will there be a return to specialization in cloud services? One-size-fits-all is a pernicious result of large hyperscaler services. This applies to skills as well as in architecture and service design. DevSecOps is the logical conclusion of trying to get everyone to do everything. What and even if AWS sees this as a major challenge or opportunity will hopefully emerge at re:Invent this week. Enterprises want to be able to start small, but be doing their cloud development and operations right and with leading practices from the beginning, without requiring an army of specialists or experts. Or making everyone an expert in everything. A tough maturity topic that should be a major discussion this week.

9:05am: Simplicity, or something I refer to as complexity management, in managing AWS resources in aggregate, especially instance types, keeps coming up. Marshalling everything required to run an HPC app can be a tall order. How AWS can streamline application use cases and dynamically bring capabilities together — and shift between them — more easily is another overarching desire by organizations. Limiting the surface area of these challenges is a key way to start slicing cloud complexity into more bite-sized pieces.

9:10am: Data sharing of real-time applications for analytics or a real-time data model is a key area of interest as well. For example, using SageMaker to run ML models on real-time data to modify digital experience on the fly is the world AWS is trying to bring to its customers. End-to-end integration is required that is also highly secure and compliant with privacy, security, and trust is neeeded as well.

9:21: Cloud sovereignty rears its ugly head. Government are busy writing specs that will have major impacts. For AWS, Local Zones like not the solution, though is likely part of it. Outposts as well.

This content was originally published here.