With every new generation of IBM zSystem family of servers, there will be an expected boost in performance, a corresponding technology dividend (e.g., reduction in the client’s IBM software invoice), and platform unique feature enhancements and extensions that will improve operational productivity and offer advanced capability. But occasionally, new functionality is introduced that overshadows these incremental upgrades, creating an inflection point in the operation of your business. Through early access to the IBM z16 platform, we have had the opportunity to evaluate new functionality and have come to appreciate how revolutionary these enhancements are.
The new functionality in the IBM z16 platform solves two significant problems related to large-scale data processing: the approaching quantum security threat and the accelerating and competitive field of AI. Let’s cover these one at a time, starting with the quantum security threat.
The Threat from Quantum Computing
With all the talk about quantum computing and how it will revolutionize the technological landscape, you may wonder where data security comes into play. In short, the threat dwarfs much of the cybersecurity breaches and ransomware attacks we hear about today. It is potentially the ultimate threat.
First, we must understand how most data is protected today, and that’s through encryption. Encryption works because it scrambles data behind incredibly complex mathematical structures that are difficult to solve, even with today’s most powerful computers. An encrypted file with a highly secure password of 16 digits would take almost 1 trillion years to solve and decrypt.
But quantum computers work differently. Essentially, quantum computing is an innovative, newly emerging technology that employs the laws of quantum mechanics to solve computing problems more efficiently. Theoretically, they can crack even the most robust encryption today very easily.
Imagine creating a master key for every encrypted file ever made since the inception of computers. Every sensitive piece of data would be at risk. Unfortunately, data breaches are commonplace today, but much of the data intercepted is encrypted and useless to attackers. But with a quantum computer (which is predicted to be available by the end of the decade), any previously captured data could then be decrypted in a “harvest now, attack later” approach, putting sensitive customer information, private medical data, as well as corporate and state secrets at risk.
Introducing Quantum Safe Computing
Due to the scope of the threat, researchers have been doing their best to develop new ways of protecting data. IBM researchers have co-developed new mathematical algorithms that are much harder for quantum computers to solve. At the forefront is developing a post-quantum cryptographic standard that will allow for a quantum-proof encryption solution. The IBM z16 is the industry’s first system based on this advancement. Our expertise combined with this new technology lets us help our clients identify the systems and data most at risk and build a plan to protect them from future attacks today.
Another notable security enhancement to the z16 is the addition of a secure boot, which makes it harder for attackers to inject code to take over the system at start-up, improving overall cyber resiliency.
Improving Decision Velocity
Let’s look at the second of these major enhancements. According to PYMNTS.com, the average consumer performs at least two transactions online daily. In the US, that is close to 1 billion daily transactions. A not insignificant number of those transactions turn out to be fraudulent, credit card fraud being the most common. Financial institutions are continuously looking for ways to better control the costs of fraud and have now turned to AI models to analyze and predict fraudulent transaction behavior patterns. These models are successful in reducing the frequency of successful suspicious transactions while reducing false positives. There’s just one problem: they take an incredible amount of processing time.
Most AI decision support models today are performed by GPUs (graphic processing units) and GPU farms. These systems perform actions requiring pattern recognition, like recognizing fraud, assessing loan approvals, route optimizations, cancer diagnosis, and more. An application will move the data from the primary processing system over to the GPU farm for processing and then move the data back. It’s this movement that introduces latency. And because of that latency, real-time processing of data and transactions is nearly impossible. Financial institutions will sample a small number of transactions in real-time to prevent fraud, which means they will miss many of them.
Imagine a surgical unit in a hospital that removes a mass from a patient and wants to have it tested. In a typical hospital, the sample is placed in a container that goes into a vacuum tube and then goes to a lab for processing. The surgical team then waits to see if it’s cancer. Imagine the same scenario, except the testing lab is right in the surgical unit. A nurse can walk the sample over, have it tested, and bring the result back to the surgeon in seconds, reducing the time a patient is at risk. This same idea can be applied to your large-scale data processing systems.
What is On-Chip AI Inferencing?
To improve decision velocity, IBM has included on-chip AI inferencing in the z16. An AI chip is a specialized chip designed to work efficiently with popular AI frameworks such as TensorFlow. Embedding it in the z16 provides advantages over traditional approaches:
- Reduced Latency: One of the foremost advantages of on-chip AI inferencing is the capability to reduce latency to one millisecond. This enables processes to be performed at the snap of a finger, which is big news for businesses that rely on speed to fulfill computing processes. For example, critical banking processes are latency-constrained due to outdated x86 GPU implementations. Typically, the inference request must be sent over the PCI bus to the GPU, and the result must be returned to the CPU. The latency introduced makes specific use cases, like real-time fraud analysis, impossible.
- Scalability: In addition to improvements in latency performance, the on-chip inferencing can scale up to 300 billion inference requests per day to drastically improve existing AI-dependent business processes and enable new business processes to be developed.
- Machine-Learning Model Compatibility: As a critical component in many AI processes and applications, on-chip AI interfacing was designed to be compatible with and optimized for common learning model frameworks such as TensorFlow, ONNX deep learning models, and IBM Snap ML.
In short, it increases the ability of businesses to apply AI processing to transactions in real-time. For example, a bank can analyze all transactions for fraud as they are happening rather than merely sampling a subset. This approach will improve the speed of all kinds of processing, enabling instant credit approvals, speeding up medical diagnosis, and better optimizing transportation networks. These are only a few of the possible use cases.
The Bottom Line: IBM’s z16 server is Worth the Migration
For several reasons, migrating to the IBM z16 platform is worth it. There are, of course, capacity improvements with a capacity improvement range of 1.12 to 1.16 based on the type of workload. There is also the standing technology dividend (Tech Dividend) that is offered by IBM to clients as they move from server generation to server generation. The Tech Dividend offers improved financial benefits relative to your current hardware maintenance and software costs.
Evolving Solutions recommends that clients act now and look for ways to protect their data from new attack vectors. Remember, your data is at risk of being stolen today with the intent of exposing it tomorrow. One way to act now is to encrypt your data using algorithms resistant to attacks by classical and quantum computers. IBM’s z16 server is the first quantum-safe system and IBM’s latest Crypto Express card available on this server offers quantum-safe APIs you can use to protect the data used by existing and new applications.
The IBM z16 server is the first system that offers an integrated accelerator that will allow you to deploy workloads that require high-speed, real-time, and very low latency inferencing. IBM’s flagship operating system (z/OS) is instrumented to leverage this accelerator, and so too is Linux. Couple this base capability with open-source tooling providing a robust AI Ecosystem and you will find yourself in a position where you truly can train your models anywhere and then deploy them on z for real-time inferencing.
As an IBM Business Partner with a z16 in our Innovation Lab, we guide you on both your AI and Quantum journeys. If you would like to learn more about the IBM z16 and the impact it can have on your organization, contact Evolving Solutions. Our IBM experts can help you define a plan and share insights from our experience with the IBM zSystem family of Servers.