Fujitsu 2012 Annual Report Download - page 61

Resource pool architecture
(2) Development of Distributed Parallel Complex Event
Processing Technology that Rapidly Adjusts Big
Data Load Fluctuations
Effective use of big data demands an ever-increasing amount of
time-series data be analyzed continuously in real time. Fujitsu
developed the world’s first distributed parallel complex event
processing technology designed for use with cloud technology
that rapidly adjusts big data load fluctuations. This research was
supported in part by Japan’s Ministry of Economy, Trade and
Industry’s Project for the Program to Develop and Demonstrate
Basic Technology for Next-Generation High-Reliability, Energy-
Saving IT Equipment for Fiscal 2010 and Fiscal 2011.
The newly developed technology enables greater segmenta-
tion during processing; when the volume of data increases, the
load can be spread across more servers, while a decrease in data
volume conversely sees processing concentrated among fewer
servers. The ability to scale event processing across servers has
resulted in a throughput function of 5 million events per second,
enabling real-time analysis of big data.
(3) Prototype of World’s First Next-Generation Server
Simultaneously Delivering High Performance and
Flexibility
The increasing diversity of cloud services—including those involv-
ing big data—has created a need for flexible systems better tai-
lored for each service, compared to datacenters for traditional
Web services.
To answer this need, Fujitsu has developed a resource pool
architecture in which the hardware components, such as CPUs
and HDDs, are linked together with high-speed interconnects.
This has enabled development of a next-generation server proto-
type that simultaneously delivers high performance and flex-
ibility. The new architecture makes it possible at all times to
flexibly configure systems to suit not just Web services and big-
data processing, but also newly emerging cloud services. Further-
more, more efficient use of hardware components minimizes
system costs and contributes to power conservation.
(4) Development of Anti-Distortion Technology to Cut
Power Used by Compensation Circuits in Ultrafast
Optical Fiber Transmission Systems
As the amount of data transmitted over the Internet swells—due
in part to growth in big data—harnessing optical fiber technology
to build ultrafast networks at low cost and with low energy con-
sumption will become increasingly important.
In response, Fujitsu has developed a digital signal processing
algorithm to compensate for waveform distortion in long-haul
transmission systems over several hundreds of kilometers. This
has enabled a roughly 20-fold improvement in compensation
ability per circuit size compared to conventional technology,
thereby extending the long-haul operating range of optical
signals. The technology eliminates the need for signal regenera-
tors in telecom carriers’ trunk-line networks and in networks
linking datacenters that provide cloud services and other ser-
vices, paving the way for ultrafast, long-haul transmission sys-
tems that are low-cost and energy-efficient.
(5) Supercomputer “K computer” Takes Consecutive No.
1 in World Ranking; Launch of PRIMEHPC FX10
Supercomputer
The next-generation supercomputer, the “K computer,” jointly
developed by RIKEN (Japan’s flagship research institute devoted
to basic and applied research) and Fujitsu, took the No. 1 position
on the TOP500 List*1 announced in November 2011 for the
second consecutive time, having previously been identified as
the world’s highest performing supercomputer in June 2011. The
K computer system is equipped with 88,128 SPARC64 VIIIfx CPUs
developed by Fujitsu to deliver high performance, exceptional
reliability, and low power consumption. The system achieved
the world’s best LINPACK*2 benchmark performance of 10.51
petaflops*3, signifying a speed of ten quadrillion (one “kei” in
Japanese; equals ten thousand times one trillion) floating-point
operations per second. It achieved this performance with a com-
puting efficiency ratio*4 of 93.2%, surpassing the previous figure.
In November 2011 Fujitsu commenced global sales of the
PRIMEHPC FX10 supercomputer, which further improves on the
supercomputer technology employed in the K computer in being
capable of scaling to a top theoretical processing performance of
23.2 petaflops.
*1 TOP500 List: A project ranking the world’s supercomputers based on comparative
performance, with rankings announced twice yearly.
*2 LINPACK: A program for measuring computer performance.
*3 Petaflops: Peta stands for one thousand trillion, or one quadrillion. FLOPS stands
for floating point operations per second, or the number of calculations the
machine is capable of in one second.
*4 Computing efficiency ratio: The ratio of actual performance reached versus
theoretically possible peak performance.
Configuration can be flexibly changed on demand
Improved usage
Improved maintenance
Disk pool
CPU pool
System A
System B
System C
High performance
059
FUJITSU LIMITED ANNUAL REPORT 2012
Performance