By Martin Banks
The High Performance Computing (HPC) division of NEC has created an interesting problem for IT managers looking for the future direction of big enterprise systems. With the recent introduction of its new TX7 family, HPC division has come up with a machine ideally suited to those IT managers’ needs, but it will not be selling any.
Confusing? Well, not really. The HPC division is dedicated to designing and building supercomputers. Its Earth Simulator system in Japan is still far and away the most powerful system yet built. This is what Jorg Stadler, the division’s European marketing manager, refers to as being “like an F1 racing car, where the cost is less important than the result. But there are a growing number of sectors—aerospace and automotive engineering, pharmaceuticals, and the like—where a more general purpose but high-powered machine is now required. This is more the equivalent of a high-end Mercedes.”
This is the market sector at which the new TX7 family is targeted. The reason it should generate some interest in IT managers providing services in general business is its specification. The machine is based on Itanium 2 processors—up to 32 of them—with Linux as the operating system. Arguably this is, in fundamental terms at least, exactly the type of specification that a growing number of IT managers will now be starting to consider as their future platform.
What does this mean to the business sector?
Stadler readily points to the possibilities of TX7 in the general business sector. “It will be running Linux for the general-purpose supercomputing and technical markets,” he said, “but it can also run HP-UX and Microsoft’s .NET environment." But, he then added emphatically, “the HPC division will not be selling it into the Windows server marketplace. I cannot say what other divisions of the company might do with the technology.”
It would be very surprising if the divisions focusing on general business sectors failed to take it—or something based on the technology—up as a business systems platform. Indeed, it is a precursor of a growing trend toward finding the solutions for future business systems in the supercomputing field.
The TX7 comes in three different chassis configurations offering up to 8-way, up to 16-way, and up to 32-way systems. The building block of the system is a processor cell coupling four Itanium 2 CPUs together with 16 GB of memory. The key component, according to Stadler, is the NEC crossbar technology used to interconnect the processor cells. This comes straight out of the development work for supercomputer systems.
The company also draws heavily on its experience as a builder of IBM-compatible mainframe systems. In particular, it exploits the partitioning skills it has gained there to provide partitioning capabilities at the cell level. This means that up to eight different virtual servers could be available within a single box.
System performance in a business environment
For IT managers, the most important factor concerning the TX7 is that it lays down a strong marker for the capabilities of Intel’s Itanium in the business server space. Intel may well have near-total dominance of the market for low-end departmental and applications servers, and the vast majority of analysts and pundits may be predicting that Intel will take over from the establishment RISC-based UNIX servers, but that might change thanks to the TX7.
NEC claims that a 32-processor TX7 yields a performance of 100 GFLOPs. This compares to an IBM Regatta system running 32 Power 4 RISC processors, which according to Stadler tops out at just 90 GFLOPs. To be fair, the GFLOP (giga floating-point operations per second) is a performance measure specifically dedicated to the needs of the scientific and technical communities, so it is not directly translatable to business applications. It is reasonable to assume, however, that with an Intel/Linux architecture, plus the ability to run HP/UX and .NET, system performance in a business environment will remain comparable.
What is more, the underlying trend in supercomputer development is increasingly toward Intel processors and Linux as the base operating system. There are already four Intel-based systems in the top 20 of the world Top 500 list of supercomputers—the accepted listing for such machines. What’s more, they are Xeon-based systems, so it is not unreasonable to assume even better performance, long-term, from Itanium-based implementations.
This is, in practice, where IT managers now need to be looking to see what systems they will be specifying in the not-too-distant future. With the Advanced Simulation and Computing (ASC) series of supercomputers at the Lawrence Livermore National Laboratory in California leading the charge toward ever-more powerful Intel/Linux based systems—and most of the leading server vendors from IBM downward building them—the trend is clear. It will be from the ranks of supercomputing that the next generation of Itanium-based, general-purpose business servers will come.
Perhaps the most important component of all in this trend is the fact that some business applications are, in practice, growing upward to meet supercomputers coming down. For example, in computing terms there is now little difference in the technologies needed to simulate fluid dynamics of airflow around an automobile body and a complex risk analysis of a multinational business venture. But now for a growing number of enterprises, the latter capability is becoming crucial.