+ What is HPC?

High performance computing is when parallel computing is not enough.

When tackling computational problems of any size, most practitioners (correctly) begin by subdividing the problem into smaller, focused components that can be run simultaneously. This is the “trivially parallel” solution and can be prototyped and/or implemented using any number of tools: Java, Python, R, SAP, or SAS.

However, trivial parallelism can only take you so far. HPC allows you to harness the full power of modern computer architectures effectively - think accelerators such as GPGPUs or the Xeon Phi - while exploiting SIMD operations and efficiently managing processor cache. You need HPC...

  • When both latency and throughput are important
  • When the inherent advantage threads have over processes is needed
  • When lock-free, wait-free, sync-free matters

+ What problem does Big Denominator solve?

Big Denominator fills the mismatch between HPC supply and demand. Those who demand or need HPC typically discover the following:

  • HPC expertise is hard to find and costly
  • Existing applications do not readily port to HPC due to legacy code base
  • Internal development of HPC expertise is a non-core business activity

Firms need off-the-shelf and affordable access to HPC tools.

+ Who should use Big Denominator?

Our customers tend to fall into any number of the following categories:

  • Face demand from management and/or regulators for more and better analyses
  • Have analysts that want to do more and better work but are computationally constrained
  • Constrain or "dumb-down" their analytics to fit within their computing capacity
  • Want to get off the "buy more servers" treadmill and, instead, get more out of each server

+ What is SIMD?

SIMD stands for "single instruction, multiple data" and is the computer science equivalent of a "free-lunch."

Multi-core processor architecture enables parallel computing, the colloquial version of which is multi-tasking. The computer can do two things at once because each core is a separate "brain."

SIMD, on the other hand, is true simultaneous computation on a single core. Under just the right conditions, the processor core can operate on multiple pieces of data literally simultaneously. The trick is understanding what just the right conditions are. Big Denominator's design patterns and tools are meant to facilitate this, nudging the user in the right direction.

The Wikipedia summary can be found here.

+ Is this anything like Apache Spark?

Big Denominator does not compete with Spark - they inhabit specific niches and can work in tandem.

Because Spark is Java-based, its raw computational capabilities and performance does not match that of a C++ based toolkit, e.g., lack of access to SIMD. However, Spark, and the ecosystem of tools that have grown up around it, offer rich access to many data sources as well as high-level data filtering and screening.

In fact, Big Denominator often uses Spark as an I/O engine, while the analytics and modeling are handled by our internal solutions.

+ Is this cloud based?

No, but it could be. Big Denominator does not host services or applications. Our tools and solutions are primarily intended for internal deployment by our customers. That said, there is no reason our tools could not be deployed within a cloud-based environment.

+ How will my relationship with Big Denominator work?

Once you reach out to us, we'll work with you to customize a package of products to meet your needs - every customer is different and deserves a custom tailored solution.

From there, we'll work with your user community and IT organizations to develop a timeline and project plan for testing and deployment. The deployment timeline is heavily dependent upon the complexity of the solution and the amount of time and resources customers can allocate.