Their next system will have a performance of over 10 Petaflop/s. That is only a factor of 100 way from Exascale, says Oniaga. He is sure that without too much changers in the design, they will get there. Again, interconnect is key.
The K computer is SPARC based with a special brand of Unix. Will that not hinder porting of applications? Was another question. The designers do not think so. Actually most of the porting work goes into parallelisation. Adapting it to the processor and operating system is only minor work.
K computer is a huge system with over 500.000 nodes. About every 15 hours a node fails. So they are working on low level rerouting of nodes so that applications do not fail.
Jack Dongarra, responsible for Linpack software used as benchmark in the TOP500, commented that to tun the Linpack to brought them on top of the list, the K system needed 22 hours. According to Dongarra it as amazing that brand new large supercomputers of this kind run for that long, right after they have been installed. The Linpack group is working on a new version of the bench mark that allows to only execute part of the benchmark, so that the run would only take about 12 hours. Of course the short benchmark should never give higher figures than the complete benchmark. With this changes the Linpack would be "Exascale ready".