In modern enterprise data centers, where flexibility is important, being able to rapidly configure or provision hardware can help prevent bottlenecks. Bandwidth demands also vary. Composability, or composable infrastructure, can improve networking speeds: data center networks must handle a variety of workloads, some high performing and some not.
Data center CPUs struggle to quickly process data, even in a composable or disaggregated infrastructure. Composability can distance resources from the nearest processing unit, increasing latency. NVMe over Fabrics can improve this problem: NVMe SSDs are incredibly fast and provide a direct path, using a PCIe bus, to a CPU. NVMe-oF makes all NVMe storage available across a network fabric.
Composability differs from hyperconverged infrastructure, another popular approach to data center architecture, in its ability to:
- Scale resources independently. This simply means that if a company needs more compute power, they can increase their computing resources while their storage capacity remains the same. This saves them money that they’d otherwise spend on storage they aren’t using.
- Access an entire network fabric, if the data center is structured that way. The network fabric brings all resources together and connects them in a software-defined pool. Hyperconverged infrastructure, in contrast, mainly focuses on virtualization.
- Dynamically provision servers and hardware, including bare metal, making them available to workloads and applications within the data center, so that workloads can be moved quickly to the best available hardware.
Problem Solving with Composability
The development of composable infrastructures indicates the need for all data center resources and hardware to be less interdependent. This shift isn’t just for storage or data centers; other technology fields are seeing it, too (microservices, for example). Tightly intertwined technologies, though helpful in limited situations, decrease flexibility, and flexibility is what most large tech companies need. In composable infrastructure, that’s the flexibility to:
- Use a variety of hardware
- Manage everything through software
- Move applications and workloads quickly, wherever they will run most effectively
- Pool resources
- Scale resources independently, so that users aren’t paying for services that they aren’t using.
Composability can also support containerization. One of the great benefits of containers is that they also run on bare-metal servers and improve flexibility and movement within the data center. Containerization is a fast-growing technology, and composable infrastructures create additional opportunity for containers to run wherever they need.
Some storage researchers and experts have questioned the long-term benefits of composability, believing it will be difficult for companies like Liqid and Fungible to fully gain traction because:
- Some businesses just don’t bother reconfiguring or reprovisioning their cheap commodity servers, preferring to toss them out after use
- Composability could be a better choice for just renting hardware temporarily, such as a colocation center or carrier hotel, but not for long-term use
However, other experts, such as storage and technology vendor Western Digital, recommend it, claiming that composability helps with data processing at the edge.
Composable Infrastructure Vendors
Some data center and service providers offer composability solutions; these vendors have features intended to solve common bottlenecks within data centers and to improve their data processing and storage capabilities.
Liqid, one of the most well-known composability providers, offers Matrix, a solution for data centers. All resources within the composable infrastructure, such as Intel Optane persistent memory and NVMe storage, connect to the network fabric. Through Liqid’s software, users can compose (prepare and deploy) servers for applications, workloads, or VMs.
Fungible, a data service startup, works to eliminate processing bottlenecks in data center networks. The company uses DPUs (data processing units) that are intended to move resources and data to the best processor and improve performance.
Hewlett Packard Enterprise’s composable service integrates with HPE GreenLake and is intended to work within hybrid cloud environments. Synergy’s HPE Composer manages the storage, networking, and compute resources. Synergy also offers virtualization and software-defined disk storage.