For me, tech conferences, such as Dell Technologies World (DTW), are places where lots of conversations happen, creating a turmoil of thoughts. This post is the result of such turmoil. It comes from the confluence of 3 different ideas and concepts and is my take on where I see compute going in the future. As with any post talking about futures, I need to preface it by saying that the content of this post only reflects my opinions and thoughts. It doesn’t reflect the strategy or roadmap of my employer, Dell.
Concept Luna
With that out of the way, let’s dive into it. The first concept that blew my mind at DTW was Concept Luna (https://www.dell.com/en-us/blog/pushing-the-boundaries-of-sustainable-pc-design-concept-luna/). The gist of Concept Luna is very simple: create a laptop that is modular enough that every component can be upgraded/replaced. As a customer, this means that once I buy the laptop, I keep the shell forever and replace the components, screen, motherboard, battery, etc… as needed. Should Dell release a new 16K screen for my 4 year old laptop, no problem, I just need to change the screen. In terms of sustainability, this is a major step forward.
This is what a Concept Luna laptop looks like:
I will wager that most people throw away their laptops because either it is not powerful enough any more, i.e. they need a new motherboard, or the battery doesn’t last long enough any more, i.e. they need a new battery. With Concept Luna, they don’t need to throw away the whole laptop anymore, just the component that need replacing/upgrading. This creates significant savings in terms of physical resources, i.e. metals and energy, consumption.
My challenge to the industry is pretty simple: Why can’t we implement Concept Luna with servers? Imagine a server where all the components can be individually replaced and upgraded. Where once I buy the shell of a 2U server, I don’t have to buy another shell every. Imagine the savings in terms of the consumption of metal resources. If I need the latest and greatest CPU from Intel or AMD, all I need to do is replace the motherboard, or have the motherboard replaced for me. This means I don’t have to migrate my application when I need to upgrade my server. I just take the server down, replace the motherboard and reboot it. Done!! Upgraded!
Event though I can go on and on about the benefits of Concept Luna for server, I am going to park Concept Luna for a minute and discuss the next concept.
Datacenter Modular Hardware System (DC-MHS)
Datacenter Modular Hardware System is an initiative by the Open Compute Project (OCP). The goal of DC-MHS is to create modular building blocks by standardizing the interoperability and providing consistent interfaces and form factor of various components within a datacenter. With DC-MHS, a server is the sum of its building blocks, with each block having a standard form factor and interface. For instance, DC-MHS specifies the size of a motherboard and the various interfaces to that motherboard. Same for power supplies, fans, accelerators elements, memory elements, etc… This is a radical change from how components are built, with each vendor having its own form factor and interface for each of the server components.
At the server level, DC-MHS brings a number of benefits to customers:
- It minimizes validation testing as the components of a server are standardized and their interfaces common.
- It helps with the availability of the industry-wide supply chain.
- It allows the delivery of faster technology and reduces the cost of transition.
- It also opens new business models, based on upgrade vs replace.
From the industry perspective, DC-MHS also offers a number of advantages:
- It reduces the barriers of adoption of new technology.
- It improves industry efficiency and innovation by having CPU suppliers to design and validate circuit boards, allowing platform suppliers, i.e. server vendors, to focus on innovation without the burden of planar design.
DC-MHS is a significant departure from how servers are traditionally designed. Traditionally, server vendors have designed their own motherboard, fans, etc… DC-MHS changes all that. DC-MHS changes the role of everyone in the supply chain but allows the server vendors to focus to be on their value-add vs the “plumbing” of a server.
Between Concept Luna and DC-MHS, you can see a theme developing, but it is still missing a key component and this leads me to the third concept.
Compute Express Link (CXL)
Since the birth of the PC and the X86-based server, there is a paradigm that has always been true: the memory and the CPU are on the motherboard. There are great reasons for this, primarily around performance, until recently, with the advent of Compute Express Link (CXL). CXL is a consortium-based initiative to develop a standard interconnect to support connecting coherent accelerators and extending system memory. To me, CXL is a game changer for the X86 space. We can now have CPU drawers connected to memory drawers connected to IO drawers, all through CXL, that can be composed based on the requirement of the application. I am not going to get deeper in CXL as I have interviewed an expert in CXL and we spent nearly 20 minutes covering all aspects of CXL. Here is the link to the video: https://youtu.be/5tnIFPmzsnM .
Putting it all together
If I put those 3 concepts together, it creates a very disrupting picture: servers are no longer this monolith, where if I want to leverage the latest and greatest CPU, memory or drive technology, I need to replace the whole server, but instead this modular concept where each component can be replaced and upgraded at will, with the shell of the server being the only constant artifact. The server is now this concept of interconnected building blocks through standardized interfaces, that can be replaced as needed.
If I take all of this one step further, each of these building blocks could include an out-of-band management module to handle orchestration and allocation of resources. With these out-of-band management modules, we now have true API-driven disaggregated composable infrastructure. This is something the industry has been talking about for a long time, but I feel that, as an industry, we needed those 3 concepts coming together to bring this idea to life.
I want to finish this post by highlighting the benefit of those 3 concepts coming together from a sustainability perspective. This is groundbreaking. Metal resources are scarce and finite. Replacing the entire shell of a server just because the need to upgrade CPU and memory is a tremendous waste of resources, so any initiative that allow either saving or reusing that type of resource is a huge deal.