Have you ever searched for a fun cat image, only to find the same image come back up again and again? You might think that cat image is popular, on multiple websites. But no – it’s the same image because the search algorithm is resetting, therefore, changing your search results.
The folks at NGD Systems know you might not be searching for cat photos, despite what Friskies wants you to believe. But as photos and videos are not only an internet concern, but also really taking a turn for enterprise IT. After all, your company might have started recording and saving your online meetings, educational videos, and hi-definition photos along with those large PDF and PowerPoint presentations, to be referenced at later times.
Not to mention the ever growing AI workplace – needing a digital assistant to turn on and off the lights. An Airport, for example, will be using similar information to try and identify people via biometrics – mainly facial recognition. To capture a face, create feature points, then search several databases for identity for a single face can be tiring. Doing hundreds or thousands of faces continuously 24/7 is something we want computers to do, without error.
Bottom line: Nobody wants a slow search with slower, repeated results and errors.
How NGD Systems Plans to Move Computing to Storage
At Storage Field Day 17, we learned about NGD Systems – a company that understands compute (processor time) is the key factor to organization and delivery. Storage can hold organized data, but no two searches are alike. If I go through my Alexa history, I might have common questions, but there are some queries that might even make your head scratch.
The idea is to simply keep the processors cooler by moving the process to storage – giving a single task to each drive – then adding drives as the database grows. The results get cataloged when new data is introduced, making the queries faster since the information is organized on a new level.
NGD doesn’t move all files, just key parts, to make the query run smoother. As new items are being introduced, the system is being auto-indexed and brought into the network. The number of process steps is reduced.
They showed an example with Microsoft called “Flash Soft”. They use FAISS (Facebook’s AI Similarity Search). But when you get to trillion item searches (or more), relevant results become harder to achieve. NGD Systems gives and end result with 4x impovement across the board. As databases grow, and more drives are added to match the data, the search time stays consistent because the drives run in parallel.
Now, I need a good plugin that will do that for finding the right tags for this article…
More in the Video
CEO Nadeer Salessi, Vladimir Alves, and Scott Shadley present this at #SFD17.