With the help of Google Africa and Intel Corporation, the iHub has embarked on a journey whose immediate realization is the setting up of a High Performance Computing (HPC) system, which we are calling the iHub Cluster.
The best part is that it was an idea was presented in 2011 from an iHub community member, Idd Salim and by December it was clear it was going to happen. It became a matter of ‘when’ not ‘if’.
It was at the turn of the Millennium when the term “Super Computer” meant a number of networked computers or a desktop computer that can achieve 1 Teraflop. Obviously we have moved on from here but the term has become superfluous in meaning. There are new terms such as High Performance Computing (HPC) have come up to mean more or less what people envisaged then when “Supercomputing” was talked about. Today we talk of computing clusters.
Supercomputers generally aim for the maximum in capability computing rather than capacity computing. Capability computing is typically thought of as using the maximum computing power to solve a single large problem in the shortest amount of time. Often a capability system is able to solve a problem of a size or complexity that no other computer can, e.g. a very complex weather simulation application. 1
Capacity computing in contrast is typically thought of as using efficient cost-effective computing power to solve a small number of somewhat large problems or a large number of small problems, e.g. many user access requests to a database or a web site. Computer system architectures, whose purpose is supporting many users for routine everyday tasks, may have a lot of capacity but are not typically considered supercomputers, given that they do not solve a single very complex problem. These are normally labeled as data centers.
The Super ComputerIn general, the speed of supercomputers is measured and benchmarked in "FLOPS" (FLoating Point Operations Per Second), and not in terms of MIPS, i.e. as "Multiple Instructions Per Second", as is the case with general purpose computers. These measurements are commonly used with an SI prefix such as tera-, combined into the shorthand "TFLOPS" (1012 FLOPS, pronounced teraflops), or peta-, combined into the shorthand"PFLOPS" (1015 FLOPS, pronounced petaflops.)
The current entry point into the TOP 500 supercomputers is 80 teraflops. The top super computer as of November 2011 is USA’s IBM built “Sequoia” computer that runs on POWER processors stands at 20.13 petaflops (20132.7 teraflops).
Academic and Communal Institution Super ComputingIn 2003 Virginia Tech crashed into the Top500 at number 7 with a super computer built if Apple Xserve G5 machines. It ran at 22 teraflops. “System X” (10) as it was known cost US $ 5 million to put together but was bringing in US $ 20-25 million annually.
From December 2011, Virgina Tech upgraded their super computer to one called “HokieSpeed” that is built of the “off-the-shelf” technology that any institution or person can quickly build. “HokieSpeed” entered at 96 on the Top500 and cost US $ 1.4 million to build.
Virginia Tech’s “HokieSpeed” have done a number of things well:
(a) It is cost effective. It has 209 nodes, each node with 2 Intel Xeon E5645 6-core CPUs and 2 NVIDIA M2050/C2050 GPUs on a Supermicro 2026GT0TRF Motherboards. All this for US $ 1.4 million for 120 teraflops.
(b) It has used standard off-the-shelf hardware and open source software
(c) It is top in the “Green List”, which is a list of super computers that are most efficient in power consumption.
So in essence for US $ 42,000, one should be able to get 6 nodes of this quality at the then prices and at least 4 teraflops. This would not get into the Top500 but all the technology would be HPC compliant and hit a number of milestones.
Placing iHub in the HPC Global CommunityDuring the research into the current HPC arena and how and what iHub can do to get into this space, a number of questions arose that would look into strategic and tactical positioning that would place iHub in good stead among the HPC global community.
- What is iHub’s ultimate goal in building the “Cluster”?
- Considering that the cluster can and should pay for itself in 1 to 2 years, what systems and personnel would be required for this outcome to be achieved?
- What benefits will the iHub community and academic institutions interested in utilizing the iHub Cluster system receive?
- Will the cluster in the future be open to serve the needs of the corporate and industrial sectors such as metrological and oil/extraction industries or international research requests from the African region and the Middle East? Considering that there is a business case in development, does the iHub have the necessary structures to cater for this?
- In order to be a respectable HPC system, one needs to enter HPC Top500 list and would need to spend around US $ 500,000 to enter the 80+ teraflops range. This figure will clearly fall in one year’s time. Is this a reasonable future aim?
Some of the answers to these questions will only be realized by actually doing this. And in the words of another iHub member, “BRING IT!”
And a more interesting question is, what would it take to break into the 80+ teraflop range and be a bona fide super computer for the iHub?