Intel has announced its Intel Cluster Studio XE tool suite to serve developers working at the heavy compute capacity coalface in High Performance Computing (HPC) cluster systems. By coalescing cluster management alongside application and data management tools in one software development suite, the company is hoping to capitalize upon its undeniably extensive work in both parallelism and multi-core computing technologies with a view to future-proofing today's systems into tomorrow.
New capabilities in the XE edition release include MPI-enabled thread profiling and correctness checking. There is also an MPI (Message Passing Interface) library that scales beyond 90,000 processes and is up to 6.5X as fast as alternative MPI libraries as measured by latency tests.
Given the "state-of-the-nation" reality we see with HPC systems having shifted to multicore nodes, there is now a logical path for these same systems to evolve to mixed multi-core and many-core nodes. As a result, the compute power of the hardware is scaling at approximately twice the rate of Moore's Law; another reality that many programmers (even those using comparatively flexible and agile development methodologies) may not have been anticipating.
Intel appears to be suggesting that HPC computing will affect all computing environments in a sort of "top down" cause and effect motion; the company says that software development solutions and tools must scale to meet developers' needs to utilize newer, faster systems.
Intel's message to HPC developers and others with Cluster Studio XE is that both node level and cluster level application performance can now benefit from a boost in terms of performance and reliability. In a single hand, the company is playing its cluster tools technology, alongside its advanced threading/memory correctness analysis and performance profiling tools, plus Intel compilers and libraries to provide that "scale to HPC on more nodes for the future" leap that it insists we now address. In the new release, the tools automatically install their remote monitoring components on every node of a cluster and monitor its internal performance. Existing MPI functions in Cluster Studio, also now included in the XE product, monitor inter-node messaging as needed.
Again with a view on future roadmaps, Intel has positioned this new release as suitable for shared, distributed, or hybrid applications. The 90,000 new processes found in the Intel MPI Library are said to be 6.5x as fast as alternative MPI libraries in latency tests. Building on this momentum, Intel reminds us that its C++ and Fortran compilers are up to 47% faster than alternative compilers. Of course there's also advanced cluster and node profiling tools for performance tuning to keep everybody smiling.
According to a pre-release product sheet, "With Intel Cluster Studio XE developers can produce code that scales on Intel Xeon Processors today while easily extending to the Intel Many Integrated Core (Intel MIC) architecture in the future." There is also, unsurprisingly (but no doubt welcome), support for parallel programming models here with commercial versions of the open source Intel Threading Building Blocks 4.0 and Intel Cilk Plus 1.1, MPI and OpenMP 3.1. Coarray Fortran is also provided with a view to preserving source code investments with minimal code changes as developers "scale forward" to many-core.
"[Developers can now] create faster and more reliable applications by finding threading, memory, and coding errors and tune performance early in the design cycle," said the company. There is also thread and memory correctness checking here with Intel Inspector XE, which is now MPI-enabled across the cluster nodes. Also included with Intel Cluster Studio XE is Intel VTune Amplifier XE for rapid node-level performance profiling so users can identify hotspots faster on the cluster nodes. Together, Intel says that these allow for testing and tuning of programs written in a "hybrid" (MPI for internode programming mixed with shared-memory models for intra-node programming) fashion.