[Connect with HPC users and learn new skills in the IBM Spectrum LSF User Community.]
It’s tempting to think of fifth-generation wireless (5G) as just an incremental improvement. 5G is the number that comes after 4G/LTE after all, but the impacts promise to be epic. 5G networks will reach speeds as high 20 Gbits per second, support 1 million devices per square kilometer, and provide 1 msec latency, unleashing the age of IoT. Everything will be impacted by 5G from high-performance cloud data centers to core and access networks to edge services and smart devices.
Accenture calculates that 5G will add $500 billion to US GDP annually(1) enabling new industries and revitalizing old ones. New applications will include autonomous vehicles, precision agriculture, and immersive entertainment. Fast, reliable communication with guaranteed quality-of-service (QoS) promise usher in new applications such as connected ambulances, robotic surgery, and widespread use of drones and automated carts for autonomous delivery applications.
Re-thinking distributed application services
In addition to new application areas, 5G will also affect existing applications and services. For applications in HPC and analytics, network capacity has historically placed an upper boundary on computing and data handling requirements. For example, an HPC application can only collect so much remote sensor data over a slow network, and a ridesharing app can only call an AI prediction service so many times per minute. With 5G networks, these constraints will go out the window, and the amount of data to be collected, processed, and analyzed will soar.
This “data deluge” will impact the full range of traditional HPC applications. For example, inexpensive IoT sensors will provide natural disaster warning systems with more frequent, higher-resolution data improving forecast accuracy for everything from extreme weather to earthquakes to tsunamis to volcanos.
A more subtle effect of 5G is that it will change how developers build distributed software systems. Today, design choices are constrained by bandwidth, latency, and cost considerations, but these barriers will fall away rapidly. Instead of a mobile app communicating with a cloud endpoint every 30 seconds, why not code a 5G-enabled app to poll the service every second? Instead of monitoring 100 real-time sensors, why not monitor 1,000 if doing so will provide a competitive or economic advantage?
The trickle-down effect of these simple design choices will be enormous. Back-end service that relied on NFS filers or relational databases won’t cut it as traffic scales to millions of writes per second. For even modest 5G applications, developers will need to embrace HPC and cloud-scale tools such as distributed file systems, key-value stores, in-memory data grids, and faster Spark-powered analytics to boost the performance of application services.
5G will depend on HPC
While current mobile networks can provide some connectivity for new services like automated cars and drones, both 5G and HPC will be essential to unlocking these next-generation services. 5G networks owe their speed to use of millimeter waves, radio signals between 30 GHz and 300 GHz. While these high-frequency waves carry more information than their 4G counterparts (4G operates between 1Ghz and 5Ghz), they have a shorter range. This means that 5G systems require much smaller cells. Whereas a single transmission tower might service a 4G service area, a 5G network covering the same area may be comprised of 100 or more small, low-cost antennas affixed to streetlights and telephone poles.
Telcos will need forward-deployed HPC capabilities to capture and analyze the vast amounts of data coming from a larger number of access points and 5G devices. For example, autonomous vehicles and drones relying on 5G services will stream telemetry to multiple 5G antennas and rely on high-performance computing, storage, and AI-powered predictive services to fuse data in near-realtime so that vehicles can operate safely and avoid collisions.
5G will have predictable impacts on infrastructure
With over 20 billion IoT devices expected by 2025(2), today’s cloud data centers cannot scale to support 5G use cases. The exploitation of 5G will only be possible with modern, HPC capable infrastructure close to the network edge.
As 5G technology is introduced, we can make some educated guesses about how data center services will need to evolve to keep pace:
- New content services – The back-end services that support new 5G applications will need to be much more scalable than their 4G counterparts. These high-performance services will likely be built using many of the same cloud-native tools and techniques used to build scalable cloud services today.
- Compute, and storage shifts to the edge – Given the real-time nature of 5G use cases, along with tiered cloud services, HPC-capable infrastructure will be needed near the network edge. To simplify management, service providers will want to manage this on-premises edge infrastructure the same way as they manage cloud services.
- Increased need for Data Management and Analytics – To accommodate larger data volumes, 5G services will need high-performance data management solutions such as HPC-oriented file systems, object stores, and distributed NoSQL datastores. Users will need tools that make it earlier to deploy and manage distributed data management and analytic frameworks at scale.
- Security at the edge becomes critical – As services migrate from well-protected cloud data centers to the network edge, security will become a critical consideration. This will be especially true as 5G networks evolve to support critical applications such as traffic control systems, industrial applications, and healthcare. Operators will need to forward deploy security technologies on par with the most sophisticated cloud operators.
For operators building high-performance 5G services, IBM offers a variety of solutions both in the cloud and for deployment at the network edge. Scalable HPC cloud services can be deployed using the IBM Cloud Kubernetes service, IBM Spectrum Computing software and IBM Watson Machine Learning Accelerator and IBM Watson Internet of Things to train and manage scalable AI services and manage IoT devices securely.
Developers can use IBM Cloud Private and benefit from the convenience and security of cloud computing from within their own data center. All of these services run on high-performance IBM Power Systems production-proven in the world’s largest supercomputers.
- How the U.S. wireless industry can drive future economic value – https://www.accenture.com/us-en/insights/strategy/wireless-industry-us-economy
- State of the IoT 2018: https://iot-analytics.com/state-of-the-iot-update-q1-q2-2018-number-of-iot-devices-now-7b/