Developers Are the Future Of VMware (Part 2): Multi-Cloud and AI
The VMware Explore 2023 keynote in Las Vegas focused on showing how VMware can help enterprises leverage the power of Generative AI to develop, operate, and optimize applications that have the potential to transform every single part of the organization, including marketing, manufacturing, procurement, legal, sales, and IT. Let’s look at a random example for the business impact of this new type of application that relies on Generative AI to complete complex ‘creative’ tasks that required significant human labor in the past.
[You can jump to part 1 of this article: Develop and Operate]
Generative AI: The Promise of Total Individualization of Content and Decision Making
The transformative power of Generative AI is based on a so-called Large Language Model’s (LLM) ability to instantly analyze vast amounts of data to automatically deliver contextually appropriate, personalized actions, content, and recommendations.
LLMs are not limited to text only, but can dynamically analyze, create, and adjust the content of video files, images, and audio.
For example, based on a web search for my name, the Generative AI may find my vacation photos and therefore it knows that I love Maine and Lobster fresh from the pound (I can only recommend visiting Young’s, but that’s besides the point). Based on this knowledge, the Generative AI can adjust storylines, images, and videos to get my attention. For example, a car commercial would use a scenic road along the coast of Maine as its setting and end with a few friends sharing Lobster at Penobscot Bay.
My friends who spend most of their vacations camping in National Parks would be targeted by a car commercial within that setting. But we can take this example much beyond the setting of these advertisements. Generative AI can adjust the actual message of the commercial to optimally fit the value of each targeted individual or group of individuals. To optimally target me, the car commercial would dynamically add some throaty engine noise and maybe an animation of how the suspension adjusts to different road conditions, and as I’m always afraid of hitting wildlife, it would show how the car’s built-in image recognition system can even evade a massive moose in pitch black conditions. This type of use case fell straight into the realm of science-fiction, only a year ago, but is possible today, provided you can afford the required hardware infrastructure.
Back-of-the-Envelope Infrastructure Cost Calculation
Staying with our example of dynamically personalized multimedia ads, the hardware cost of researching, generating, serving up, and storing a two minute video ad could be significant. Targeting a single person could require roughly 2 hours of processing time on a high end system that costs around $30,000. Therefore, if we optimally sequenced this workload, we could create targetd ads for approximately 12 people per day, using our $30,000 investment. Of course we could store and re-use these commercials on similar targets, but the point remains that if we want to target tens of thousands of potential customers on a continuous basis, infrastructure cost would be significant. This is where VMware’s multi-cloud management stack wants to come in to save the day.
In part 1 of this article I looked at Tanzu Application Platform, VMware’s developer platform based on the Backstage open source project and at Tanzu Application Engine that provides self-service application spaces for developers to provision, independently of the underlying infrastructure. Now let’s drill down one level deeper into Tanzu Hub, VMware’s multi-cloud management platform that translates resource requirements into infrastructure code, and ultimately, provides the desired infrastructure.
Tanzu Hub: Knowledge Is Power, GraphQL Is Knowledge
Tanzu Hub is based on a central GraphQL database that collects and consolidates operations data across data centers, private and public clouds, and edge locations. Instead of accessing each individual environment separately and writing cloud-specific queries, users can write a single query to access entities of app environments across AWS, Azure, GCP, and vSphere. Coming back to our targeted advertisement example, we could now search for underutilized resources to process these videos. The chart shows the user ask a question in plain English (“Show me all of my VMs that have GPUs and show an average utilization of under 50%”). The Tanzu AI Assistant translates this question into a GraphQL query and delivers the answer back to the user in a the form of a single API response that can include resources from AWS, Azure, GCP, vSphere, etc. The user then wants to find out if there are specific times where these resources are most underutilized (“Show me GPU utilization of these 3 servers over the past week aggregated by hour”). The AI Assistant remembers its previous response, queries for the same 3 servers, and delivers the answer back to the user who can then schedule her video processing workloads on these resources.
This shows how the Tanzu Hub GraphQL API and the AI Assistant allow users to define queries in plain English to find temporarily underutilized cloud resources that they can take advantage of, instead of purchasing additional capacity.
Generative AI Writes Infrastructure Code
The same Generative AI-driven assistant can translate plain English requests into infrastructure-code for Pulumi, Terraform, etc. specific to Azure, AWS, GCP, vSphere etc. This makes the deployment and operations of app stacks across different clouds consistent, as the AI Assistant takes on the translation work, while the user simply specifies a generic set of requirements. Thinking this further, the user could simply say: “Provision the infrastructure needed for me to analyze and process 50 hours of 4k video in 8 hours.” The AI will then analyze and calculate the server configuration (CPUs, GPUs, RAM, storage, IOPS) needed to get this done. Then I could tell the user the estimated cost for running this infrastructure on each individual cloud and let the user decide where to deploy.
External apps, such as observability platforms and, of course, VMware’s own developer portal (Tanzu Application Platform) can use Tanzu Hub’s GraphQL API for root cause analysis, cost optimization, preventative maintenance, budgeting, observability, what-if scenarios, and capacity management.
One Search For All Cloud Infrastructure
Tanzu Hub’s GraphQL API combined with AI Assistant also enables users to ask general high level questions and receive actionable insights within the overall organization context. Some examples:
“What actions can I take to increase application performance without adding cost?”
“What would it cost if I stretched my Kubernetes cluster across multiple availability zones?”
“What apps need my attention right now?”
“What’s the potential business impact if my cloud migration fails?”
“How can I lower cost, while preserving compliance”,
Or you can make more specific requests:
“Create a new cluster for our ERP software on the East Coast.”
“Add 24 CPU cores and 128GB of RAM to my EKS cluster.”
Answers Within the Business Context
As a platform engineer or infrastructure admin I can ask the Generative AI about the business impact of certain actions or the business risk of the failure of a certain set of cloud resources:
“What applications and business KPIs will be affected if the disk on this host server fails?”
“What is the business risk of an in place software upgrade?”
“Are there any compliance concerns regarding the use of the latest Ubuntu operating system images for our PostgreSQL clusters?”
“What are the steps of a controlled shutdown of the following 3 servers?”
This Time, It’s Personal
While the Generative AI considers historic context data when offering insights, it also remembers an individual user’s preferences and past decisions.
For example, I may have shown a clear bias toward AWS resources over Google Cloud or vice versa or I insisted on fixing a specific app’s database latency way before reaching a critical threshold or I may have shown a preference for NoSQL databases over SQL, whenever possible. The Generative AI considers these past decisions, including their impact, for future interactions.
Final Thoughts
Coming back to the original question, asked in the beginning of part 1 of this article: “Is VMware the right horse to bet on for my cloud native application strategy?” Both parts of the article showed a lot of evidence that VMware actually “gets it.” The company has executives in place who understand customer requirements (they do have 500k existing customers after all) and who are in the process of rearranging VMware’s product portfolio accordingly. I generally like the story they are telling and the product demos that seem to back these stories nicely. But can they execute and will they be able to execute under the roof of Broadcom? This is a question that I refuse to answer at this point, but when we look at the alternatives we find that there is no such thing as a ‘sure bet’ when it comes to app transformation.
