Big picture
The cloud era started with salesforce.com (1999) and AWS (2006).The first cloud was built on a regionally centralized cloud, with data centers in key regions (North and South America, Europe, Middle East & Africa, Asia Pacific)
Centralization of the cloud enables economies of scale but is ill-suited to support latency and bandwidth requirements of certain applications.
We need a new "global-local cloud" that will provide “local delivery” while maintaining global reach.
This trend is already underway with content delivery networks (CDN).
The new global-local cloud will allow the optimal placement of applications by considering the needs of the services. The global-local cloud will be self-learning, automated and self-optimizing digital fabric for the new era.
Latency
Latency is a major issue for content delivery. This is a direct result of the limitations imposed by TCP.Maximum throughput (bits/second) = TCP Window Size (bits)/ round-trip latency (seconds)
With a typical TCP window size of 64 kilobytes and a round trip time of 100 ms, the TCP throughput is clamped at 5 Mb/s. The throughput requirements for a 1080p stream and 4K are 6 Mb/s and 15 Mb/s respectively.
If latency is reduced to 20ms, the throughput can be improved to 25Mb/s. CDN networks have responded to these facts by providing distributed architecture.
Akamai, a CDN network, operates 170,000 caches in 1300 public and private networks in 102 countries.
Large content providers have started to establish commercial agreements directly with local and regional network operators. Bell Labs Consulting predicts that an increasing number of global providers will leverage the proximity of local service providers. It is estimated that by 2025, 60% of cloud servers will be located in edge locations.
Dynamic Scalability
The virality of social media has led to the dynamic popularity of certain content. The "Budweiser Puppy Love" commercial spread on social media and received 13 million views on just the first day.
Cloud services must be able to handle this highly dynamic traffic profile, which requires advanced monitoring, automation, and orchestration. SDN solutions will enable the creation of fully dynamic data centers. The tools and technologies needed to build this new cloud are still beginning to emerge.
Future
By 2020, the cloud will transform into a true hybrid of global and local resources. The localization of the cloud will be driven by two factors. First, the explosion of the unicast streaming video. Second, the emergence of hyper-local, high-performance and low-latency enterprise applications.
According to Nielsen, traditional TV viewing is gradually declining while unicast streaming services like Netflix are increasing. The network architecture needs to shift focus from broadcast delivery of content to on-demand unicast delivery.
There is also a new need of doing the local processing of high-bandwidth streams for video analytics. Real-time analytics requires sending many unicast streams across the network but in the opposite direction from the streaming video delivery for entertainment. Localization of the cloud resources at the edge of the network allows processing to be located near clusters of devices, reducing network traffic while maintaining the benefits of centralized analytics.
The global-local cloud can decrease the latency to single-digit millisecond range enabling a new class of applications that require tight feedback control loops. One such application is permitting humans to interact with virtual objects rendered in the cloud as if they are physically co-located. Today such capabilities are only possible through the use of significant intelligence in the endpoint devices.
Both virtual machines and containers will address different virtualization needs and will continue to play an important role in the local edge clouds and data centers.
Intelligent automation and orchestration based on the past and present traffic patterns would be an important factor in the new clouds.
Networking in the cloud will be enabled by software-defined networking (SDN). Typically, a network service is comprised of multiple functions known as a service chain. In the global-local cloud, functions in a service chain may be distributed across the cloud.
The new business model
The global-local cloud will provide low-latency services with global access to users regardless of their location. There will be global service providers (GSPs) and local service providers (LSPs). The relationship between GSPs and LSPs may follow one of the four following models:LSPs provide infrastructure for GSPs
LSPs provide hosted as-a-service function to global cloud services
GSPs provide local infrastructure
LSPs provide global services
By 2020, it is likely that we will see a mix of all of these scenarios. However, GSPs will be of a limited number (say 10) because of the cost involved in building a global infrastructure.
By 2020, the majority of services must globalize their reach but localize the delivery to be acceptable for a highly mobile consumer base. For high-performance and mission-critical services, local delivery will become essential. The end result of this revolution will be a creation of an eco-system in which a handful of major global-local service providers leverage the resources and services of 100s of local-regional service providers.
Comments
Post a Comment