IBM and Exascale Computing: the Biggest of Big Data
IBM has won a 5-year contract to explore and develop technologies for the Square Kilometer Array radiotelescope. The effort will push the boundaries of computing in all directions.
In cooperation with the Netherlands Institute for Radio Astronomy, known as Astron, IBM (sponsor of Business Agility) will establish a research center in The Netherlands to explore technologies capable of reaching the "exascale" in computing. This is in service of the SKA program, an initiative to build a continent-spanning radiotelescope by the year 2024. Hundreds of researchers across twenty nations are involved in the planning.
The telescope will be constructed either in western Australia or in South Africa; the location will be selected this year. Its approximately 3,000 dish antennas will span 3,000 kilometers (the graphic represents the way the antennas will be distributed). The total area of these dishes amounts to nearly a square kilometer, hence the name. The collecting area of this instrument will be more than 13,000 times larger than that of the Aricebo telescope in Puerto Rico.
The scale of the data moving and processing required by the SKA is on the order of 100 times what is required by CERN's Large Hadron Collider -- and instead of being bursty like the LHC data, the SKA's will be more continuous. Each of the SKA's thousands of dishes will produce 160 gigabytes of data per second. The total daily data generated by the project will be around an exabyte -- 1018 bytes -- which is twice the amount of data that moves across the Internet worldwide today. That data must be reduced and processed in real time. The project will permanently store around an exabyte of data per year. Nearly 50,000 miles of fiber-optic cable will be needed to move all the data around the telescope array.
The processing required at the center of the SKA will be around an exaflop per second. This is more than 13 times the total processing power of the 500 largest supercomputers operating on Earth in November of last year. According to this poster on the Top 500 site, the collective power of the top 500 should reach the exaflop level by 2016, and by 2020 the single fastest supercomputer will reach that level on its own.
The SKA computing facility will need to be greener than any that exists today. Current data center technologies expend 98 percent of their energy moving data around and only 2 percent processing it, according to InformationWeek.com. IBM will be exploring three-dimensional chip technologies in which the processing elements are orders of magnitude closer together in order to reduce power consumption.
For working storage, the leading candidate is phase-change memory, which is faster and more durable than flash; and it doesn't lose data when turned off, resulting in lower power consumption relative to DRAM.
Overall, the computing center's power consumption must beat today's standards by a factor of 10 to 100, according to Popular Science -- not only to meet the massive scale of the requirements, but to save energy on cooling in the southern desert.
Permanent data storage will need to be on tape. Rotating magnetic media would be far too expensive, and tape storage is projected to reach the level of 100 terabytes per cartridge within 10 years.
The Square Kilometer Array project will stretch computing in all its dimensions, and the results for the world outside of observational astronomy will be greener data centers comprising faster computers connected by faster networks. Everyone's agility will benefit.