Reconfigurable Critical Operations Processors (RCOP)
What
is it?
The
RCOP performs several functional tasks.
It acquires large amounts of real-world streaming or packetized
signal data (analog and/or digital measurements); performs an often
complex set of operations on the data in a defined (e.g. real-time) time
interval; and communicates the results of the operations for further
processing, display, or control of some external elements. The design of
such a processing system involves resolving the conflicting factors of raw
performance versus flexibility.
The
Reconfigurable Critical Operations Processor has three elements: 1) a
General purpose PC with large
memory and disk storage and high speed access to an external wired or
wireless network; 2) a Reconfigurable processor based on the Xylinx 10
million gate class FPGAs; and 3) a high speed data acquisition interface
specific to the particular source of signals.
High performance is insured through the combination of on-board
DRAM and SRAM with the FPGAs. Specialized
high performance I/O interfaces and closely coupled I/O cards can enhance
the performance.
In
addition to high performance, reconfigurability is provided through a
library of previously tested and debugged “processor images.” These
images can be rapidly downloaded to the FPGAs to configure the processor
to optimally process a specific class of signals from a specific
instrument or source of data. Alternatively,
the stored images can provide a number of possible approaches to a
specific source of data.
Numerous,
tests by various groups from NASA, the Air Force, and MIT’s RAW project
at CSAIL have demonstrated that reconfigurable processors based on FPGAs
can deliver anywhere from ten to more than five hundred times the
performance of optimized software {C code} running on a single Pentium IV
class general purpose PC depending on the balance between I/O and
computing that is required for the specific benchmark that was being
tested. The FPGA has an
advantage both when it comes to tasks requiring a large amount of I/O with
relatively small amounts of computing and tasks that require large amounts
of computing that can be effectively parallelized in the FPGA.
Classes
of Problems
Reconfigurable/FPGA
computers are not suitable for all problems.
They shine, however, at problems that require repetitive
calculations and quick software changes.
Some of these are:
- Genomics
research
- Pharmaceutical research
- Medical
Imaging
- Molecular
modeling
- Pattern
matching
- Communications,
Image Analysis, Anomaly Detection
- Radar
signal and data processing
- Modeling
and simulation
- Weather,
Financial
- Forensics
- Security, Cryptography
Because
FPGAs run at “the speed of silicon”, complex algorithms can be
accelerated to run hundreds of times faster than on a P4.
FPGAs make “real time” attainable.
If
you are presently using or contemplating using grid or cluster computing
to solve a complex algorithm or compute-intensive task, you may find an RCOP
to be a more time and cost effective alternative. Please
call us at 508-299-3897 or drop an inquiry to mrlewis@datadefense.com
to find out more.