Hoopoe™ – GPU Cloud Computing

[toc]

Introduction to Hoopoe™

Hoopoe™ Logo
Hoopoe™ is an innovative, pioneer, GPU Computing (GPGPU) cloud infrastructure and service.

Hoopoe™ was designed from ground up to deliver faster than real-time processing performance using latest GPGPU technologies and architectures.
With a flexible design and deployment model, it can power private (internal) clouds or serve as hosted service with public cloud providers, such as: Amazon™ AWS™, Microsoft® Azure™ and more.

Why is Hoopoe™ so unique?

It is the first and only infrastructure to deliver faster than real-time distribution and data processing performance
It is the first and only commercial grade, cloud scale, infrastructure designed from ground up to deliver true GPU Computing capabilities, using latest technologies: NVIDIA® CUDA® and OpenCL™

Built with cross-platform tools and able to use cross-vendor GPU frameworks, Hoopoe™ allows using of GPU/CPU hardware for computational intensive tasks, while running on existing systems.

Cloud Service Motivation

The introduction of GPU hardware for compute applications opened the way for performance improvements not expected before (according to Moore’s law with a traditional CPU). A single GPU device can deliver up to several TFLOPS (1000 billion floating point operations per second) and this number is expected to grow linearly in the near future.
For comparison, one will need a cluster with 10’s of multi-core CPUs (excluding additional server hardware and communication interconnects) to reach the same of performance. Therefore, combining many GPUs together, it is possible to get very high processing speeds, at much lower cost.

GPU hardware is relatively simple and not complex when compared to a CPU, as it allows only arithmetic and computation logic to be used, making it ideal for scientific or industrial processing needs.

After some years that GPUs are being used for compute purposes around the world, it is clear that not every organization is capable of handling them. A great level of IT expertise is needed to maintain a performance system, select hardware components and create an efficient environment.

In such cases, cloud services are perfect fit for small business organizations (SMB/SME) to get started and avoid high management costs, in the long-term and hardware costs for initial steps.
Enterprise organizations, on the other hand, face a different problem, how to scale up when needed, but having limited resource in house. This is where cloud services provide the answer – the ability to scale with no overhead, in addition to maintain a budget wise status with pay-per-use.

Features

Real-Time Performance

Hoopoe™ core engine was built for faster than real-time distribution & data processing performance. It can handle millions of different tasks per second using sophisticated, patent-pending, algorithms.
This allows highest possible resource utilization and efficient processing.

Security

Using unique abstraction mechanisms and patent-pending technology, Hoopoe™ is the only infrastructure that provides a truly secured, isolated, execution environment for GPU hardware and Fusion/APU/CPU processors.

Scalability

Incorporating sophisticated core engine for distribution & management, Hoopoe™ can handle millions of tasks per second and a similar amount of notifications.
In addition, it can scale without overhead from few compute resources (GPU/CPU) to thousands in real-time.
These features are not provided by any cloud or cluster infrastructure.

Cross-Platform Execution

Hoopoe™ was designed with most advanced, standard, cross-platform frameworks, to deliver high performance environments.
In addition, support is provided for multiple operating systems (both clients and server) to address variety of needs:

Microsoft® Windows® (Desktop, Server, HPC Server)
Linux
Apple® MacOSX
Oracle™ (Sun) Solaris™
Embedded profiles and variants

Latest GPU Computing Technologies

Support most recent GPU computing software frameworks: NVIDIA® CUDA® 5 and OpenCL™ 1.2.

Support most recent GPU/CPU hardware architectures: NVIDIA® (all generations, including Fermi™/Kepler™), AMD and Intel. Fusion and APU processors are also supported to allow true heterogeneous computing.

Standard API

Hoopoe™ exposes standard, simple, web service API for management, control and general administration of compute resources/facilities.
API interface complies with SOA guidelines, so multiple systems/technologies can communicate with Hoopoe™, whether written in Java, Python, Ruby, Perl or else.

Advanced Storage Solutions

Hoopoe™ incorporates an internal storage engine called HoopoeFS™, providing a very simply and intuitive interface for storage tasks, and ideal solution for scientific applications.

In addition, it can connect with any storage provider, public (Amazon™ S3™, Microsoft® SkyDrive™ Google Drive and much more!) or private (internal file-systems).

Value-added Apps (GPU accelerated)

FFT – FFT routines (single/double precision)
BLAS – Basic Linear Algebra routines (single/double/complex precision)
LAPACK – Additional Linear Algebra routines and solvers (single/double/complex precision)
Encodeo™ Video Processing – Analytics, decoding/encoding (MPEG-2, MPEG-4, VC-1, H.264 etc.), enhancement
Image Processing – Common operators or complex algorithms (machine vision, enhancement etc.)
Support for all media types and highly extensible

Target Applications

Hoopoe™ targets the following applications, but very open to address new fields:

Big-Data, Minining and Classification
Digital Signal Processing
Image Processing & Enhancement
Video Processing / Transcoding
Medical Imaging
Remote Sensing
Scientific Research / Simulations
Finance
Geophysics – Seismic Processing