The data center network is on the verge of one of its most dramatic upgrades in bandwidth in decades. Virtualized data centers, as well as hyperscale and rack scale environments, are driving the need for higher bandwidth networks. These environments have the computing power to host dozens of virtual machines per server and analyze massive amounts of information in near real-time. The storage systems that support these environments can meet the computing demand, thanks to high-performance all-flash arrays. Connecting the compute layer to the storage layer is the network and right now, for many data centers, it is the bottleneck.
The bottleneck and perceived networking complexity lead many hyperscale and rack scale environments to attempt to leverage direct attached storage (DAS) so that the storage network is not as critical to operations. Even virtualized infrastructures are trying to build alternative networks with hyper-converged infrastructures (HCI).
The problem is DAS brings with it, massive capacity inefficiencies and resource allocation issues. In fact, several studies have shown that most hyperscale and rack scale environments use less than a third of storage capacity. Given that current capacity is increasingly flash based, it is an expensive waste of resources to use only a third of it.
The Answer is More than Gbps
Most networking vendors are now providing faster bandwidth; Fiber Channel (FC) is delivering speeds of 32Gbps, and Fibre Channel over Ethernet (FCoE) is delivering 40Gbps. While the bandwidth increase is essential, IT planners need to look for more.
First, IT planners need to understand the total throughput of the switches they purchase. The switch needs to support full line rate so that the data center doesn’t face switch “traffic jams.” Additionally, not all applications need access to the same levels of bandwidth but many applications, thanks to virtualization and containers, share the same connection. To prioritize bandwidth properly, the infrastructure needs to support Quality of Service (QoS).
Second, IT planners need network infrastructures that help them do their job better. Troubleshooting modern environments is difficult because of the amount of abstraction and the sheer number of connections. Modern networks need to provide deep insight by using analytics to analyze telemetry data from dedicated ASICS integrated directly into the switch.
Third, IT planners need network infrastructures that can move beyond decades-old storage protocols and integrate new protocols while providing simultaneous support for the old ones. Simultaneous support enables the organization to move to them gradually as they refresh the compute and storage layers.
As an example, IT planners should prepare for NVMe. NVMe is a protocol designed specifically for flash. It provides a higher number of commands and deeper queue depth than legacy SCSI. It also has a network variant, NVMe over Fabrics (NVMe-oF) which promises the same low latencies for arrays as internal storage provides today.
The problem is the protocol first appeared as an option for the server, then for storage arrays. These first use cases helped system-level performance replacing legacy serial attached SCSI (SAS) but once the data leaves the server or the storage array, old SCSI latency comes back into play. Switch vendors have added support for NVMe and are providing dual (SCSI and NVMe) support. Now, viable host bus adapters (HBAs) that enable an end-to-end NVMe connection, are appearing on the market. The result is latencies that rival DAS storage, which means that virtualized, hyperscale and rack scale environments can get the performance of local storage with shared arrays while keeping the efficiencies of a shared solution.
There is no question that many data centers can take advantage of more network bandwidth but bandwidth is only one reason to upgrade the network infrastructure. Modern network switches provide intelligence in how bandwidth is provisioned and utilized. They also support new, advanced protocols like NVMe. In our blog, we cover another reason to upgrade the network, more rapid deployment of network infrastructure.
To learn more about the next generation of networking, join Storage Switzerland and Cisco for our webinar “Faster, Smarter, Simpler – The New Requirements in Storage Networking“.