Networks have historically been constructed using a hierarchical topology of rack and row switches connected into large ‘directors’ at the data center core. This architecture generates a number of ‘hops’ as packets traverse the network from edge to core to edge, being handled by switches along the way, each adding latency to the transaction. Fiber Mountain has a new software-defined networking solution that promises to reduce that latency, and lower cost and complexity over traditional core-edge networks.
Instead of a typical network interface card (NIC), Fiber Mountain uses high-density, MXC optical connectors in each server, which connect into the Alpine Connect Panel (a patch-panel type of device) at the rack or row level. The routing intelligence is maintained in a software layer called the Alpine Orchestration System (AOS) that runs on a separate appliance and manages packets as they’re sent between servers directly, distributing the network function.
By eliminating multiple switch hops and flattening the network stack with Fiber Mountain’s Glass Core technology, applications can see latencies of less than 10ns, typically, rather than the 500ns that are common for the industry’s fastest switches. This 50x per-connection speed increase and single hop topology can result in an order of magnitude better performance, compared with the typical multi-switch network.
On the cost side, fewer switches means less infrastructure to buy and less infrastructure to maintain. This translates into lower CapEx for network hardware and cabling, lower OpEx for power and cooling and a reduction in administrative overhead. In addition to cost savings, the performance boost Fiber Mountain’s technology gives the network seems to be well-suited for environments also looking at software-defined storage.
Improves Software-Defined Storage
Historically, the network has been the weak link in many software-defined storage applications that abstract server-side storage resources. Moving large amounts of data between high performance servers has been a problem with traditional networks. Decentralizing the network intelligence and moving it into an independent software layer may make this technology an ideal fit for software-defined networking.
A few years ago we were introduced to “I/O Virtualization” (IOV), a technology that promised to simplify networking hardware in the data center by replacing dedicated NICs, HBAs and aggregation switches with a rack-level IOV switch and virtual interfaces in each server. This technology was interesting but never offered enough advantages to really catch on.
Fiber Mountain’s software-defined networking takes that concept several steps further, beyond just increasing network flexibility. By potentially replacing traditional switch infrastructures it decreases cost even more, but adds another valuable benefit – lower network latency. By reducing latency a couple orders of magnitude and driving infrastructure costs out of the network, this technology can potentially open up another strong use case for aggregating and abstracting server-side storage, also known as “hyper converged storage”.