Partnership with MS, developers produces array of services aimed at broad variety of compute-intensive workloads
Accelerating analytics of all kinds is the driving force behind a partnership between Intel, Microsoft, and the developer community that is yielding blazing fast servers that easily scale, according to this story by Slashdot Media Contributing Editor John O’Donnell
John O’Donnell
Demand for ever-more-sophisticated analytics is driving research in to more efficient software, but may also upend a rule of system design that has been part of big data since from the very beginning.
Developers such as Intel Corp. are incorporating machine learning and other advanced techniques with the potential to reduce processing time by helping applications decide for themselves what options to ignore based on the results of previous calculations.
Intel is also applying its latest processors to servers optimized for analytics that could reduce the complexity and total cost of big-data hardware implementations by packing far more power into a smaller space than ever before.
Intel has already gone into great detail on the design changes to its latest family of Xeon E7 chips that optimize it for compute-intensive big-data analytic applications.
Earlier this year Intel and Microsoft collaborated on a white paper that set new records for performance of Windows Server 16 running on a quad-processor Intel Xeon E7 v3 system that was able to run complex queries on a 100TB dataset — the equivalent of a database with 600 billion records in its largest table — in just 5.3 seconds.
Fast hardware for complex analytics
The effort was part of an ongoing partnership between the two companies to optimize the cloud-enabled Microsoft SQL Server 16 databases for compute-intensive applications on Intel platforms.
The speed of the single system was largely due to Intel’s Data Center SSDs and 6TB of DDR4 system memory, both of which helped keep much of the data in memory and avoid I/O bottlenecks, according to a white paper describing the system and the tests.
Intel went further last week, releasing results showing a performance increase of 36 percent it attributed to its switch from the four Xeon E7-8890 v3 processors from the earlier benchmark with four new E7-8890 v4 chips.
Upgrading the SQL Server 2016 columnstore engine to the latest version of AVX2 instruction delivered performance numbers 2.3 times higher than the previous test; running queries in parallel rather than sequentially added another 25 percent, according to Intel’s announcement of the test.
The point was not simply to brag about enviable performance numbers on a TCP-H test. The demonstrations were a continuation of Intel’s effort to give customers reference frameworks and guidelines showing how to restructure traditional scale-out big-data analytics clusters to reduce total cost of ownership, complexity and effort.
Advantages of scale-up as well as scale-out
Unlike traditional scale-out clusters, scale-up designs running recent generations of Intel Xeon processors are flexible enough to be configured and reconfigured to run varying types of workloads. Regardless of the purpose, properly configured, the new systems will deliver top performance at lower overall TCO, than previous generations, according to an April announcement describing three sample frameworks: one designed for maximum throughput, one that balanced compute power and throughput and one designed for maximum compute capacity.
A previous Intel report on scale-up vs scale-out performance showed a 334-processor system running Xeon E7–4890 V2 clusters could deliver more than 2.5 times the performance of a 1000-node Intel Xeon E5-2680 cluster, and cut four-year maintenance- and operational costs by 37 percent.
Cloud computing has stretched the definitions and specifications of HPC applications so far it’s difficult for hardware engineers to even imagine keeping up with chipsets that can’t be reprogrammed, according to Kushagra Vaid, Microsoft’s general manager of cloud computing division, who was quoted in an April EETimes story about the impact of reprogrammable FPGA processors on datacenter servers and workloads.
Intel has updated its list of servers designed for analytics and Windows Server 16; it has also published examples of server products from Intel partners that are optimized for specific types of workloads.
Some apps can run almost entirely in the I/O network, others are very compute- and processor-intensive, Vaid said. That turns the flexibility that comes from features like the FPGA chips Intel is designing into future generations of Xeon and Xeon Phi processors into a critical element of systems design for both traditional HPC applications and increasingly demanding analytics.
Continuing expansion of demand for analysis
Globally, the market for sophisticated, big-data business-analytics market will grow by approximately 50 percent between 2015 and 2019, from $122 billion to $187 billion, according to a May report from IDC.
Nearly half of the $55 billion IDC predicts users will spend on big-data software will to end-user query, reporting, analysis tools and data-warehouse management, according to the report.
During the same year, spending on hardware for high-end analytics will reach $28 billion, reflecting both the scale of the data and solutions customers are demanding, but also the scale of the IT resources required to satisfy them, the report said.
“Organizations able to take advantage of the new generation of business analytics solutions can leverage digital transformation to adapt to disruptive changes and to create competitive differentiation in their markets,” said Dan Vesset, group vice president of IDC’s Analytics and Information Management group and lead author of the study.
“Data analytics is one of the fastest growing areas in technology,” according to Mike Pearce, Ph.D. HPC Developer Evangelist at Intel, in the announcement of analytics reference frameworks from Intel partners. “We strive to continually innovate and improve the capabilities of our flagship Xeon processors; at the same time, through close partnerships, we team with a robust ecosystem of software providers across the world to expand platform features and optimize their applications to deliver cutting-edge business analytics services.”
The post Intel, Microsoft Team Up for Screaming-Fast, Scale-Up Xeon Analytics Server appeared first on Go Parallel.