Teradata Corp. (NYSE: TDC), the big data analytics andmarketing applications company, today announced engineering advancements to the Teradata Database that deliver analytic performance and system efficiency through new memory and CPU optimizations.
These enhancements strengthen Teradata’s approach to in-memory computing and enable customers to seamlessly and automatically realize the greatest benefit from their investment in memory.
"Teradata is relentlessly dedicated to engineering a smarter, simpler way to leverage memory and CPU to drive performance," said Scott Gnau, president, Teradata Labs. "Blindly throwing additional memory at a problem has diminishing returns, particularly when it comes to big data. Teradata’s sophisticated approach automatically and efficiently places the right data in-memory to get the performance they need for the best cost."
In Teradata Database 15.10, Teradata Intelligent Memory will be enhanced with the following capabilities, which reduce the load on memory bandwidth, improve CPU efficiency, reduce I/O to disk, and improve overall system efficiency:
Query pipelining and new in-memory table structures
Pipelining is an innovative approach to query processing where the output of one query step feeds into the input of the next query step without leaving memory. This greatly increases system efficiency and throughput by reducing unnecessary data movement. Also, while this data is held in memory, it will be stored as column-partitioned tables, instead of row-partitioned tables, which reduces the memory footprint and speeds consumption by the CPU.
Exploit CPU instructions and cache
New algorithms take advantage of Intel’s vector instructions and on-board cache to apply operations in parallel. This increases CPU throughput and efficiency and further reduces the amount of data moving in and out of memory. The Teradata Database will also optimize Intel’s new Haswell processor, which will feature additional vector instructions, further improving CPU performance, and efficiency.
New data temperature measurements
Teradata Virtual Storage now uses new weightings for measuring how frequently data is accessed by distinguishing between tactical and strategic workloads. Tactical workloads, which are generally high-priority business queries, increase the heat of data faster than strategic workloads. The hot or frequently used data is moved into memory faster. This aligns data in memory with business priorities. Teradata also increased the precision of the hot data stored in-memory by allowing logical I/O’s, data referenced in-memory, to also count towards data temperature. The result is even more accurate alignment of business needs with data held in-memory to improve analytic performance.
These enhancements to the Teradata Database will be available in the first half of 2015.
