Breaking Down Those Walls: How CXL Could Reinvent the Data Center
By Raj Hazra, Micron
The data explosion intensifies the demands on compute infrastructure and data center. We are generating and analyzing more data than ever before, with the growing ubiquity of AI workloads and application-heavy devices – which will only expand with the deployment of 5G. While this transformation has the opportunity to provide new insights and a competitive advantage, the architectural foundation that has driven data center computing over the past two decades is challenged to keep pace with these load demands. sophisticated work.
To meet the needs of these data-rich workloads, it is essential to design a flexible and scalable platform architecture. Future data centers need heterogeneous compute, redesigned memory and storage hierarchy, and open, agnostic interconnect to tie everything together and enable composable systems that can scale with workloads.
Move to a heterogeneous IT world
Heterogeneous computing has been around for decades. GPUs have worked in tandem with CPUs since the early 2000s to give a boost to graphics-intensive workloads like gaming. But the expansion of high-performance computing applications and the rise of AI have resulted in the need for additional computing resources. To power these new data-centric workloads more efficiently, companies are combining various types of CPUs, GPUs, AI accelerators, and FPGAs (or collectively, XPUs) that work together to do the heavy lifting for the specialized needs of the industry. every application.
Using a range of XPUs in a heterogeneous environment opens the door to solving complex problems more efficiently and easily. The challenge is to connect these disparate compute and memory resources to an interface that can keep pace. Today, CPUs rely on a variety of interfaces for communication between processors. GPUs are often connected via PCI Express, memory is connected via DDR channels, and the list goes on. The move to heterogeneous computing will require moving some of these interconnects to a higher performing industry standard interface allowing new features such as memory tiers, pooled memory, and even memory and storage convergence. And to unleash architectural innovation and choice, we need an open standard widely accepted by industry.
CXL unlocks the architectural freedom of the platform
Enter the Compute Express Link (CXL). CXL is an open interface that standardizes high performance interconnect for data-centric platforms. .
This platform architecture freedom is why we think CXL is so important, not only because of its capabilities and performance, but also its ability to connect accelerators and multiple tiers of memory and memory. storage to the processor. It is a necessary step forward to create more composable systems with attractive memory and storage hierarchies and meet the various workloads and business needs that require a balance between performance and capacity, such as:
- Provide a memory-intensive application with more capacity with memory tiers to improve overall performance, such as a mix of low latency direct connect memory and higher latency large capacity memory
- Improve virtual machine density by allowing a cloud provider to host more virtual machines by having more memory capacity, directly connected as well as via CXL to the processor
- Hosting large databases with a caching layer provided by storage class memory to improve the performance of the in-memory database, giving it a much larger memory footprint
- Improve enterprise-class infrastructure efficiency with the use of CXL-based attached memory and storage
And, above all, CXL is an open standard. Any company can join and contribute to the CXL Consortium, which includes all the major companies in our industry. At Micron, we are thrilled to be deeply engaged with other industry leaders who are leading the way towards a CXL compatible future, enabling an open and advanced ecosystem.
This collaborative innovation is essential to build flexible and optimized infrastructures for the data center. The biggest challenge I hear talking to CIOs is predicting the infrastructure and capabilities they’ll need for ever-changing workloads. In this rapidly changing technological landscape, these needs are difficult to predict, but unfortunately the current status quo forces IT teams to be deterministic in their architectures. CXL solves this challenge by allowing the platform to compose systems and create optimized infrastructures as you go, providing the flexibility to scale with changing business needs.
Predictions for the future of CXL
Given this potential for activating the restructured data center, I have a few key predictions of what CXL will deliver:
- CXL will activate the memory areas. CXL will allow architects to create large pools of volatile and persistent memory through a direct, high-speed interface. Memory of all types will span multiple infrastructure pools and become a shared resource.
- CXL will scramble the lines of storage and memory, providing more powerful computing hierarchies. CXL will open up possibilities for inventive new types of memory and storage that don’t fit perfectly into traditional categories. You will no longer think of “DRAM for memory and NAND for storage”. Instead, there will be mixed uses for the two, and the memory-storage hierarchy will be advanced with emerging technologies such as storage-class memory.
- CXL will be ubiquitous. As a truly open and ubiquitous interconnect, CXL can go beyond data center and cloud platforms to edge-based platforms. This ubiquity will fuel emerging use cases in AI inference, in-memory databases and more. From the edge to the cloud, the industry will innovate by combining unique solutions for these emerging workloads.
As CXL expects widespread adoption in the data center over the next several years, new ways in which servers, storage, and networking are architected will make 2021 archaic. You’ll feel like you’re looking at a car from 50 years ago, thinking, “We made cars like this?” CXL will deliver ultra-high capacity memory and greater bandwidth, unify the data ecosystem, and unleash limitless architectural freedom and innovation – ultimately reinventing today’s data centers and meeting the demands of today’s workloads. work focused on the data of tomorrow.
By Raj Hazra, Senior Vice President and General Manager of Microns IT & Networks Business Unit