Driven by the relentless and increasing volume and velocity of data, there has been a renewed interest in using FPGAs to execute operations in parallel for the real-time performance needed to drive decision making in the enterprise. The operators of the world’s largest data centers and leading search engines are accelerating their data analysis and search operations using FPGA-based systems that dramatically speed the entire lifecycle of data analysis from ETL to indexing, search and business intelligence.

So why did it take so long for FPGAs to catch on in the datacenter? FPGAs, when not properly architected, are rife with complexity, which has traditionally stalled their use. However, enterprises have reached their breaking point in the battle for extracting meaningful and actionable intelligence from their data. The rapidly growing volume and velocity of data coupled with the increase in security threats mean there must be a dramatic shift in how things are done. The slow status quo is no longer acceptable.

What once could be solved using spare hardware and open source systems now needs the greater horsepower, lower latency and ability to instantly analyze disparate types of data that only new, purpose-built hardware can provide. Take fuzzy search and matching, for instance. Searches on inexact data typically take many hours to complete and data must be indexed to support each individual field to make it faster. What’s more, most tools only enable you to search with an edit distance of two which isn’t nearly enough to catch the majority of misspellings, abbreviations and other inconsistencies common in today’s data. FPGAs changed all of this.

FPGAs are enabling new types of servers to execute massively bitwise operations in parallel. These analytics servers, like the 1U Ryft ONE, can replace 100s of contemporary servers while delivering instantaneous insight into both streaming and batch data—even correlating the two—all while providing massive TCO savings.

The “properly architected” piece of the equation has been easier said than done until now. What we saw at Ryft, as have others in the industry, is that there must be a significant shift in data analytics architectures in order to keep up with the torrent of data.

In our next post, we’ll be talking more about the benefits and future of FPGA-accelerated systems. To learn more, visit “Can FPGA-accelerated Systems Eliminate x86 Bottlenecks and Speed Value From Data?

2 responses to “Data Center Trends: FPGA Acceleration is the Answer to Enterprise Data Analysis Performance Challenges

  1. I reached to the same conclusion independently just half hour ago.. 😀 Is there any existing solution that combines FGPA and data structures to pull the requested data out of Huge DB?

    1. Our Ryft ONE utilizes a heterogeneous computing environment to include FPGA technology and allows users to specify structured information (if they so choose, or you can search unstructured) in what we call record definition formats (RDFs). This is managed by a front-end linux environment with an open API. You can freely download our open API _here_

      I’m not precisely sure whether or not the Huge DB being referred to is this one or not:

      But if it is, and assuming you have access to that database and can load it onto your system in any exportable format that may be available to you, such as perhaps XML, JSON, or even plain unstructured, then it would be trivial to load it onto a Ryft ONE and search it at rates up to 10 gigabytes per second (yes, that’s gigabytes, not gigabits.) In fact, that’s true of any database at your disposal: if you have a way to export it and load it into Ryft ONE, you can search it really fast.

      There are other at-least-partially-FPGA based solutions out there as well on the market, and more and more are popping up every day. Large companies including the likes of Microsoft are making use of FPGAs with proprietary and/or modified FPGA development boards, and have even written whitepapers on the subject. One of the attractive things about our Ryft ONE device is that it perfectly abstracts its heterogeneous computing techniques (to include FPGAs) behind an open linux-based API so that the hardware acceleration is completely abstracted away, allowing the end user to interact with the system via any programming language (such as C, C++, Java, Python, R, Scala, etc), shell scripts, command lines, SQL, clustered architectures (such as Apache Spark) or even REST APIs. This opens up accelerated data analytics to occur at any point in a network – be it in a datacenter or at the edge – for everyone, whether they are programmers, analysts, data scientists, and even non-technical types.

Leave a Reply

Your email address will not be published. Required fields are marked *