Reduce cloud latency for remote employees and offices

Break down cloud latency

Cloud latency is a matter of distance and equipment. For example, the latency from California metro areas to a Google Cloud data center in Los Angeles is under 10 milliseconds when using a direct path, according to a cloud networking study by ThousandEyes, a network intelligence company. In contrast, connections from the Washington metro area to that same Google Cloud data center suffer six times the latency, while intercontinental users of U.S. cloud systems suffer delays of hundreds of milliseconds.

It should also be noted that the ThousandEyes measurements are under the best of circumstances, with measuring agents at internet service provider hubs in various cities. In practice, organizations often use suboptimal broadband services to connect remote sites into an enterprise WAN and the internet and latency to cloud infrastructure can be much higher.

Cloud latency for remote users has long been a problem for enterprise WANs. Network performance to locations such as retail stores, branch offices and warehouses has become crucial to the digitization of business processes, decentralization, streamlining of corporate structures and adoption of cloud services.

In response, companies should apply software defined networking (SDN) to WANs, known as SD-WAN, and adopt cloud network services to help mitigate the disadvantages of remote seclusion.

The basics of SD-WAN

Long before organizations considered displacing enterprise systems with cloud services, many companies had already consolidated their IT infrastructure to a few central locations to improve efficiency, versatility, reliability and security.

These distributed organizations with consolidated IT faced the same problem as today's remote cloud users -- how to deliver enterprise network performance and reliability in under-served locations. The distributed organizations ultimately applied SDN to quasi-consumer broadband and wireless network service, which became SD-WAN.

SD-WAN evolved from a combination of WAN optimization products and SDN technology, as vendors adapted a software control plane to manage the various types of wide area circuits and treated them as a virtual network. Over the last several years, enterprises have widely deployed SD-WAN as a way to centrally manage WAN links, while lowering cost and improving reliability through the use of multiple broadband or wireless circuits. SD-WAN has seen significant industry consolidation as SD-WAN specialists have been acquired by integrated technology providers like Cisco, HPE and VMware.

SD-WAN offerings include the following core features, most of which improve network performance, reliability and latency for both enterprise and cloud connections.

• A circuit-agnostic software control plane that can centrally manage any type of WAN access technology. This includes frame relay, cable or DSL broadband, wireless broadband or satellite links. A unified management console controls all WAN circuits in an enterprise.
• Circuit bonding technology to create virtual WAN connections from two or more physical links. The control plane then monitors and optimizes packet flows over the various links to provide quality of service, data compression, protocol acceleration, load balancing and caching.
• Enhanced reliability using various error correction techniques.
• End-to-end security using IPsec tunnels.
• Extensibility and management automation via REST APIs.

Together, these technologies can deliver multiprotocol label switching equivalent performance and reliability over consumer-grade services or high-latency connections such as satellites. The performance improvements of using SD-WAN for remote connections will vary by application and by the type of circuits provisioned, but users can expect double-digit percentage reductions in packet loss, jitter and cloud latency.

SD-WAN will improve all remote connections, whether to a central data center or cloud service. To work properly with cloud infrastructure, an enterprise needs virtual endpoint appliances in its cloud environment to work with its other corporate termination points, whether those are data centers or remote offices. There's also network acceleration software specifically designed to work with the major cloud providers that can increase data throughput, reduce latency and cut data traffic.

Bringing cloud networks to the edge

Remote locations that are underserved by traditional wireline services have continually been problematic for enterprises. However, recent enhancements to satellite network services to include Geostationary (GEO), Medium Earth Orbit (MEO) and, in the future, Low Earth Orbit (LEO) satellites can deliver high-speed connectivity anywhere.

For cloud users, a critical missing piece had been the connection between a satellite provider and cloud network. The task has recently been simplified via partnerships between Microsoft and three satellite providers. ExpressRoute for satellites directly connects ground stations to Microsoft's global network through private links to provide predictable latency. By next year they plan to use SES MEO satellites that provide up to 10 times the throughput of older-generation technology.

Cloud providers are also partnering with wireless carriers to incorporate cloud infrastructure and network points of presence in cellular base stations and switching hubs. For example, AWS Wavelength puts AWS storage and compute resources inside a provider's 5G network to deliver sub-10 millisecond latency to users near what AWS calls a Wavelength zone. The service will be available on Verizon's 5G network in 2020, but AWS is working with other wireless providers to expand availability.

When available networks aren't enough

Sometimes, no amount of connection bonding, data compression and protocol optimization is enough to move the required data to a location or application in a reasonable amount of time. In these situations, you can copy data onto a portable storage device and ship it to the cloud provider.

Products like AWS Snowball, Azure Data Box and Google Cloud Transfer Appliance combine a portable hard drive with software and cloud services to facilitate mass data migrations for DR and content delivery. Options like Snowball Edge and Azure Stack Edge add limited compute resources to allow local pre- and post-transfer data processing. These devices come in ruggedized housings that make them suitable for remote worksites with limited network capacity that also lack environmentally controlled computer rooms.

One increasingly common cloud use case for remote locations is IoT data analysis in which devices, such as Snowball, are used to seed a large volume of historical IoT data to the cloud. With that said, a limited-capacity remote network can accommodate the incremental stream of sensor data.

Alternatively, for IoT applications generating high volumes of data, the edge device can be used to store and preprocess data for about a month at a time before sending it to the cloud data center for further analysis.