Thanks for printing! Don't forget to come back to Trenton Systems Inc. for fresh articles!
High Performance Embedded Computing
What is High Performance Embedded Computing (HPEC)?
High Performance Embedded Computing, or HPEC, is an off-shoot of the more generic High Performance Computing (HPC) term that has been around for years. HPC at its most basic, can be described as the aggregation of computing power into systems that can achieve higher performance than a standard workstation. What makes HPEC unique is that an HPEC system offers a long-life, high performance computing system based on long-life components such as embedded processors, chipsets, network interface controllers (NIC), PCI Express (PCIe) switches, video cards, power supplies, etc. This long-life system design approach in high performance embedded computing provides a stable HPEC hardware platform suitable to system applications that require long field deployments, and need a stable hardware platform over the lifetime of a project that the software developers can rely upon. High Performance Embedded Computing, or HPEC hardware platforms are desirable becasue that can virtually eliminate expensive operating systems, component driver, and application software compatibility issues caused by random, and oftentimes unannounced hardware platform and system BIOS upgrades.
High Performance Embedded Computing, or HPEC systems are designed with an eye toward applications that require the computing of vast numbers of small problems for a long period of time, (also called high-throughput computing) or a single large problem more quickly than would be possible with an off-the-shelf computing solution. Generally, achieving HPEC goals involves the grouping of individual processor boards within a common rackmount computer enclosure and/or individual computers themselves, known as “nodes,” into, “clusters,” which are capable of communicating over an interconnect either amongst themselves or with a master controller. This description is purposefully vague because High Performance Embedded Computing encompasses such a broad gamut of technologies, communications protocols, software tools, architectures and project goals.
Common applications of High Performance Embedded Computing include the Scientific and Applied Research fields, Geological Exploration, Big Data, Finance, Statistical Modeling, Battlefield Surveillance, Telemetry, Medical Diagnostic Imaging, and Industrial Automation. A Trenton Systems whitepaper entitled, “Implementing High Performance Embedded Computing (HPEC) Hardware,” explores the implementation issues in more detail for HPEC systems.
High Performance Embedded Computing (HPEC) Technical Challenges
A typical high performance embedded computing hardware solution comprises a system host board (SHB), backplane and rackmount computer system designed exclusively to solve the data throughput needs of high density embedded computing applications. These high-end embedded applications require robust computing performance, so the SHB and system-level technologies must be able to take full advantage of the ever increasing number of processor cores and native PCI Express 3.0 links available in today’s high-performance processors. Taking this embedded computing hardware design approach can result in a system solution that delivers a 5x increase in aggregate system bandwidth with lower data latencies.
Correctly routing PCIe 3.0 single traces is a design challenge that few companies can handle well, and taking short cuts in system host board/single board computer and backplane designs which utilize the PCI Express 3.0 interface will always result in suboptimal system performance. For example, with all previous generations of PCI Express it was best practice to keep traces well below 16 inches to insure optimum performance, but the PCIe Gen3 specification makes the length requirement even more restrictive. The Gen3 specification requires a pre-validation of the link before any data transmission, and if the automatic equalization training cannot establish a reliable link, then it won’t allow the transmission of data at 8Gb/s speeds, resulting in slower than expected card-to-card as well as card-to-SBC data transfers within a typical system design.
These considerations are critical in the design of a backplane, such as the HDB8259, which supports multiple Gen3 high-performance I/O cards;including as many as four, x16 PCIe 3.0 GPU accelerators such as NVIDIA Tesla K80 cards inside a single HPEC system.
High Density Embedded Computing (HDEC) Products Drive HPEC Systems:
Trenton’s HDEC Series of system host boards, backplanes and integrated systems drive many of today’s High Density Embedded Computing systems. For example, today’s HEP8225 system host board supports your choice of two, long-life/embedded Intel Xeon E5-2600 v3 or v4 processors and a total of 80 native PCI Express 3.0 links that are routed to edge connectors for use on an HDB82xx backplane. Future HDEC Series system host boards will support 88 native PCIe 3.0 links when the next generation of high performance or server-class Intel Xeon processors become available. The promise of HPEC systems all come together with a selection of highly configurable 2U, 4U and 5U HDEC series rackmount computer systems.