For several hundred potential investors gathered in London’s Genesis Cinema earlier this year, the London round of Techstars 20151 gave them an opportunity to see the latest technology ideas put through their paces. Sure enough, there were some interesting demos, and polite applause rippled around the recently-refurbished multiplex. That is, until the big screen filled with Google’s global map. The ensuing simulation of 100 million dynamic objects being identified, sorted and pin-pointed by their GPS co-ordinates in under five seconds drew audible gasps.
The sheer speed of the operation – 15 times faster than any comparable database technology – was certainly breath-taking. That it was conducted entirely in the cloud added to the realisation that this could be a real money-spinner: an innovation that was infinitely scalable without degradation or spiralling running costs. As such, it would be capable of handling the staggering data volumes that drive the Internet of Things – a veritable Google of Big Data.
For those with £2.5 million to invest in its further development, GeoSpock (www.geospock.com) had placed itself, quite literally, on the map. The company and its slick GeoWarp core technology is initially targeting a DBaaS market in the geospatial sector that could be worth £650 million a year. Its wider potential, e.g., in biomedical, financial, communications and other Big Data processing applications, could be worth £24 billion, while the projected Internet of Things could give it a head start in a market worth £1 trillion.
As GeoSpock co-founder and CEO Steve Marsh says, scalability is the key. “Database solutions were originally designed to handle data that rarely changed, and even newer databases, such as NoSQL (Not Only Structured Query Language) solutions, struggle to keep up with the rate at which data is now being generated.” For example, in the field of telematics, fleet operators now have up to 100,000 moving sensors generating data every second – that’s about 3.2 x 1012 (3.2 trillion) datapoints over the year that are constantly being refreshed. “It imposes an inevitable trade–off between scale, data throughput, and real-time responsiveness, with performance the usual victim. At GeoSpock we’ve eliminated that compromise,” says Marsh.
The breakthrough stemmed from research into another fiendishly data-intensive environment: the human brain. The topic proved fertile ground for Marsh whose PhD thesis in Computer Science came to grips with the neurons, synapses and glia that communicate and process up to 400 billion bits of information per second. And with each of the brain’s 200 billion neurons connecting to up to 200,000 others, the number of possible data pathways is, according to one neuroscientist, greater than the number of stars in the entire universe!2
Mapping where data clusters and diverges in this dynamic process is critical to understanding the cause and effect of many physiological conditions. For Marsh, simulating neural network transactions in real-time led him to create HiveMind3, a super-compiler for the scalable custom Bluehive super computer hosted by Cambridge University’s Computer Architecture Group.4
Unlike a conventional compiler that translates program statements into machine executable object code, a super compiler selects and optimises the code for the intended domain-specific usage. The process achieves consistent sub-millisecond simulation times that can be maintained for larger networks by adding more machines. It is these key insights that inspired the GeoSpock architecture … one that is able to maintain real-time performance, regardless of the size of the database or how often it changes.
In early 2013, the potential of Marsh’s research caught the eye of Dr. Darrin Disley, a life scientist with an enviable track record in backing and growing promising start-ups. Together, the duo set up GeoSpock, and embarked on a round of pitches and demos that garnered some enviable accolades, including a coveted listing in Business Weekly’s 2014 Killer50 rankings of disruptive technologies; a Cambridge University Young Entrepreneur of the Year Award, and a Tech Nation pitch to leading investors and policy-makers at Number 10.
It also enticed another Cambridge scientist-cum-entrepreneur, Dr. Jonathan Milner, to reinforce his early investment in the business and join the board as a non-executive director. “The super smart boffins at GeoSpock have cracked the problem of scaling geolocation. The results will be disruptive and amazing,” he promises.
By the close of 2014, the business had raised more than £700,000 in seed funding, had entered the prestigious TechStars accelerator programme, and started deploying the GeoWarp engine to power a variety of next-generation extreme-scale, location-based applications in the mobile social networking, geospatial and logistics space.
Currently, the company’s developer-friendly REST API gives users the ability to create, update and search GPS-tagged objects on the GeoSpock server. These objects, referred to as ‘locatables’, have four characteristics:
- A unique ID, either specified on creation by the client or generated by the server
- A type, specified by the client and expressed as an integer. Types can be used to partition locatables into multiple collections
- A geospatial location (GPS co-ordinate), specified as a latitude-longitude floating point pair (in degrees)
- As application-specific data, specified as a flat JSON object where every value is a string
Upon a user-generated request, locatables are retrieved within latitude–longitude bounding boxes using ‘k-nearest neighbor’ searches. Particular attention has been paid to security for the cloud-based service, with every endpoint in the GeoSpock API being https encrypted and with private API keys being employed for client authentication.
Following a successful transatlantic trip to Esri’s annual user conference, the GeoSpock team is busy back in Cambridge and at its new hot desk at the Geovation Hub in London,5 refining its development model, and assessing the response to its Series A funding round. With the advent of billions of devices constantly generating data, and with businesses seeking to harness and make sense of it, this is one development that looks set to travel at warp speed!
2.Neurons, Synapses, Action Potentials, and Neurotransmission, Robert Stufflebeam, Consortium on Cognitive Science Instruction. (www.mind.ilstu.edu/curriculum/neurons_intro/neurons_intro.php)
3. HiveMind: Providing an efficient programming model for extreme-scale real-time neural network simulation. Steven Marsh. University of Cambridge Computer Architecture Group.
Subscribe to our newsletter
Stay updated on the latest technology, innovation product arrivals and exciting offers to your inbox.Newsletter