New generations of applications and workloads like AI/DL (Artificial Intelligence/Deep Learning), and HPC (High Performance Computing) are breaking the seams of entrenched storage infrastructure models and frameworks. We cannot continue to scale-up or scale-out the storage infrastructure to meet these inundating fluctuating I/O demands. It is time to look at another storage architecture type of infrastructure technology – Composable Infrastructure Architecture.
Infrastructure is changing. The previous staid infrastructure architecture parts of compute, network and storage have long been thrown of the window, precipitated by the rise of x86 server virtualization almost 20 years now. It triggered a tsunami of virtualizing everything, including storage virtualization, which eventually found a more current nomenclature – Software Defined Storage. Both storage virtualization and software defined storage (SDS) are similar and yet different and should be revered through different contexts and similar goals. This Tech Target article laid out both nicely.
As virtualization raged on, converged infrastructure (CI) which evolved into hyperconverged infrastructure (HCI) went fever pitch for a while. Companies like Maxta, Pivot3, Atlantis, are pretty much gone, with HPE® Simplivity and Cisco® Hyperflex occasionally blipped in my radar. In a market that matured very fast, HCI is now dominated by Nutanix™ and VMware®, with smaller Microsoft®, Dell EMC® following them.
From HCI, the attention of virtualization has shifted something more granular, more scalable in containerization. Despite a degree of complexity, containerization is taking agility and scalability to the next level. Kubernetes, Dockers are now mainstay nomenclature of infrastructure engineers and DevOps. So what is driving composable infrastructure? Have we reached the end of virtualization? Not really.
Evolution of infrastructure. Source: IDC
It is just that one part of the infrastructure landscape is changing. This new generation of AI/ML workloads are flipping the coin to the other side of virtualization. As we see the diagram above, IDC brought this mindset change to get us to Think Composability, the next phase of Infrastructure.
What is Composable Infrastructure?
To think about composability, we have to think about disaggregation first. There was a picture posted in one of SNIA® webinars from years ago that has stuck with me that perfectly explains disaggregation.
Virtualization and Containerization vs Composability (Source: SNIA)
If Virtualization or Containerization is to combining different functional parts (compute, network, and storage etc) to make a platform for applications and workloads (A&W), composability is going the other direction, where the functional parts (compute, network, storage etc) are dynamically assembled and orchestrated to meet the needs of the A&W. When the A&W needs and lifecycle change, these functions are dynamically reduced, disassembled even, and reassemble with different ratio and weightage to meet the new workload demands. That is the beauty of composability; the wonders and the powers of Composable Infrastructure.
Being pedantic, I looked up many definitions of Composable Infrastructure. This is the one I like best – “Composable infrastructure brings together compute, storage and network fabric into one platform, similar to a converged or hyperconverged infrastructure. It also integrates a software-defined intelligence and a unified API to “compose” these fluid resource pools.” (quote: IT Pro Today)
Compute eXpress Link (CXL)
At the heart of the shift to disaggregation is CXL. Widely regards as the standard of low level, high performance, ultra low latency interconnects, CXL has an bright and exciting future. The earliest iteration of CXL is about composable memory parts, where high bandwidth memory resources can be composed and allocated on-the-fly to serve CPU processors, especially the ones with high core counts to keep them as busy as ever. One of the best videos I watched online about the prowess of CXL in memory composability was a video released by Liqid 2 weeks ago. The video is inset here.
Of course, there are more “functional parts” coming into CXL. In the next few years heading to the end of this decade, GPU, DPUs, xPUs, and more using the PCIe ecosystem to bring forth CXL Fabrics. There is a great video by SNIA® introducing CXL Fabrics in version 3.0 inset below.
Composability fitting for AI/DL
The AI gold rush is happening now. The new wild, wild west is upon us, suddenly jolted to life with the introduction of ChatGPT 3.0 in November 2022. Generative AI is on everyone’s lips now and there is a mad, mad rush to starting training and building the next big thing. Large Language Models (LLMs) became the transformative frameworks for Generative AI, and to get there, the infrastructure and the frameworks that support these infrastructure have to be extremely dynamic, flexible and agile to meet the ever changing workloads of AI/DL.
Legacy infrastructure architectures would not cut it to meet the ever fluctuating networking and storage access patterns and delivery of AI/DL workloads. Thus a Composable Infrastructure Architecture would be fitting for these kinds of AI/DL applications and workloads. For instance, some parts of the applications during the development, and the production and the lifecycle of AI/DL applications are better suited to be offloaded from the CPUs would find itself given the right amount of functional resources at the most optimal time, whilst the platform balances its disaggregated resources in the most efficient ways. Perhaps during the training periods of AI/DL applications, more storage resources are added to address the exponential write I/Os and message passing and updates to the learning model to develop its accuracy and correction. But in the production phase of the AI/DL applications in the platform, the demand for write resources and networking are lesser now, having shifted the demand to reading resources and engaging the high volume of users querying the AI applications for accurate answers and predicated actions. And composability is able to make all this happen on-the-fly seamless, without stopping the applications to transfer from one state to another, and reapplying new learned details from new interactions to enhance the applications in the AI platform.
Summing it up
It is still early days. The understanding of Composable Infrastructure Architecture only belongs to a cohort of vendors and passionate technologists. As AI/DL workloads are pushing the brackets that break the legacy infrastructure platforms, we will come to a point where disaggregation and composability are a necessity for applications that deal with voluminous petabytes and exabytes of data, those demanding near instantaneous answers from consumers of these applications globally. What I see now is, and where it is going are what I learn from the experts in the industry every day.
I often used SNIA® as a reference point for anything storage, because I have been involved with SNIA® in different capacities for almost 17 years. SNIA® CSTI (Cloud Storage Technologies Initiatives) involves itself in educating industry about Composable Infrastructure Architecture, promoting its adoption and integration.
I also follow GestaltIT‘s podcast, Utilizing Edge, where CXL initiatives and industry news are often discussed by luminaries and experts in the field. Check out those podcasts in the earlier seasons discussing about CXL, where many technology vendors are converging to put together the building blocks for Composable Infrastructure Architecture.
The road ahead is exciting.
