Access VoltDB's latest eBook called SQL VS. NOSQL VS. NEWSQL – FOR TELCO.
First, can you tell us more about what VoltDB does and how?
VoltDB is an in-memory data technology platform that facilitates data-driven AI. So maybe Cloudera or Snowflake or whatever engine runs the machine learning and generates a model that you import into VoltDB, then run it as part of your transactional processing. VoltDB is about bringing all those elements together, to execute accurate, automated, actionable decisions in under 10 milliseconds (ms).
A good example is a Chinese multinational vendor that supplies fraud management software. By running our technology on its real-time data platform, this company enabled a customer to apply the 1,500 rules that are necessary to determine if a credit-card transaction is fraudulent or not. As a result, that customer is able to reduce fraud by 83% while increasing the capacity to handle transactions by a factor of 10.
This is because we can handle millions of transactions per second and our decision response time is in single-digit milliseconds. Users run VoltDB to run their rules codified within VoltDB and can operate at high scale and super low latency.
The usual alternative is to kick off a project with a bunch of developers to build code, then implement it and return the machine-learned results. Typically, by the time you're done interpreting what you have discovered from the process and operationalizing it, that learning is already in the archival store.
How does the technology play into monetizing 5G?
5G is going to almost mandate bringing that level of intelligence to a family of use cases closer to the edge. The two kinds of edges that have the highest likelihood of becoming key to monetizing 5G are on-prem[ises], and network edge.
The network edge is provided by the mobile network operator, and we can think of examples like AT&T and Azure working together in the US to bring Azure edge zones to market. Likewise you have Verizon working with AWS Wavelength which is essentially a colocation data center with Verizon’s network data center and they call this the network edge.
The idea is you'll have six data centers in a city, say, and you probably have a 10ms round trip to them. Right now, operators worldwide are working to bring that time down from 20 or 30ms to single digits. We’re asking what is the point of a low latency ping round trip if the value creation in the middle of that trip takes an exorbitant amount of time? If value creation is slow, there’s no value in it.
We have low latency communications, we need to explore how that affects operators’ digital transformation and their customers’ transformation? How does it help industrial automation and the enabling and usefulness of digital twins, for instance? How do we join all these dots together to provide value towards automation?
What are we talking about in terms of time regarding value creation?
A single millisecond round trip with 500ms to create value is too slow. The value creation is in the automated decision-making, which, depending on the situation, can involve personalization, business process, decisioning, real-time control feedback, and maybe telemetry data.
If you’re using telemetry data to run a wind farm and something goes wrong, you need to act fast to power a windmill down before the torque does any damage. That’s what you want rather than collecting all the data then trying to decipher what happened. In real life, the windmill carries on, the torque builds up and could shear the aluminum tubing that houses that windmill. With fast decisioning, you can prevent it.
Timing can be critical regarding customer experience: We work with a customer value management (CVM) vendor on what started as a technology optimization journey, but we discovered the hidden value of sub-10ms decisions. It turns out that when a customer contacts a mobile network operator, there is a budget of 250ms for the round trip, but a window of only 4 to 7ms to present the best-option offer to them. That offer is based on call detail records and other customer data. The rest of the time is taken up assembling the artefacts to render what the end user needs to see on their screen for them to tap the button and accept the offer. It is literally the blink of an eye in which the customer looks at their screen and makes their decision.
By getting the best-option offer to customers in no more than 7ms boosted the CVM’s acceptance rates for offers by 253%. The big lesson here was that the nature of the offer and the propensity to purchase need to coincide exactly for the subscriber to take it up. That's an example of the hidden value of enabling accurate, automated decisions in under 10ms.
What do you think the main areas are for this super-fast decisioning combined with low latency communications?
Enterprises need a system of multiple entities working together as a business process subsystem: They need to take streams of data from a number of relevant components and make immediate sense out of it. So far, this has proven very hard to do. We’ve been talking about extracting massive benefits from big data analytics for years, but the German word ‘Verschlimmbesserung’ sums up progress for many. It translates as making something worse by trying to make it better.
Applications that leverage VoltDB could be in supply chain management or authority shipping management, container organization, or securing asset tracking. Business operations and automation, and securing your assets are not two different things; they need to be integral to each other.
Security has to be intertwined throughout your entire business operations. It cannot be a separate thing that's running on the side. That means you need the business logic to run your business optimally, and that business logic includes focusing on one thing, while preventing someone breaking into the system elsewhere. One measure to address this is only allowing agreed permissible communication protocols, not anything else.
You can incorporate all this into your business logic for day to day operations or millisecond to millisecond operations, it just requires the platform that can bring both flavors of intelligence to work together. And if you if you miss either of them, if you miss the window of opportunity around an event happening, you're behind on at least one of them.
Can you talk a bit more about the potential regarding security?
We’ve worked with an ISP in Japan which used VoltDB to prevent distributed denial of service (DDoS) attacks on their customers’ sites. It has a 100% record so far in stopping them before customers have been affected.
In the US, there is huge fraud from ad bots which steal from content publishers. The publishers bid for real estate slots on websites, depending on who’s looking at any given moment and the information provided about that person. Ad bots sell content publishers spots on fake websites and present them with traits of non-existent people. One of our customers has been able to tackle this problem which has proved difficult previously because it all happens in milliseconds.
In my opinion, we should not approach this from the point of view of a handful of use cases, but as a capability that can be applied to low latency communications, then pick a sector, any sector, and apply your imagination.
Download VoltDB's recent eBook called SQL VS. NOSQL VS. NEWSQL – FOR TELCO.