IBM Targets In-Memory Databases With Power8
IBM and SAP have joined forces to add the HANA database to IBM's Power platform upgraded specifically to run in-memory analytics.
The partners said SAP HANA would run across the entire Power8 processor line. IBM's E880 server based on the Power8 processor brings with it 200-Gb/sec memory bandwidth, "intelligent" data caching and 16 TB of storage capacity, yielding an aggregate 3.2 Tb/sec of bandwidth, IBM said.
SAP HANA was recently tested on the E880 server running "very sizable" in-memory analytics workloads, achieving what the partners claimed was much faster access to HANA's in-memory database.
IBM stressed that the Power8 line of processors and servers were specifically designed to accommodate huge increases in the size of datasets and databases. Those growing volumes have pushed many workloads onto in-memory databases, necessitating huge memory capacity.
In an attempt to anticipate what has come to be known as "big data," IBM equipped the Power8 line of processors with a proprietary L3/L4 cache memory subsystem to handle greater data volumes running on in-memory databases like SAP HANA. "We knew we were taking a risk," Brad McCredie, vice president of IBM Power Systems Development, noted in a blog post announcing the deal with SAP HANA.
IBM further claims that Power8 cores outperformed Intel's Haswell EX processor cores while running SAP HANA analytics workloads based on SAP's mixed workload benchmark. Along with performance improvements, McCredie claimed the Power8 line could scale to higher core counts. That should translate into the ability to scale servers to consolidate workloads while maintaining large datasets as a way of reducing the latency of database queries.
Along with intelligent data caching, IBM touts Power8's multithreaded computing capability and high bandwidth memory capacity. Combined, those features yielded a four-fold performance increase over x86 platforms, it claims.
The partners added that efforts to tune in-memory database performance on Power8 would include taking further advantage of SIMD (single instruction, multiple data) technology and transactional memory, a feature used in multicore chips that IBM first introduced commercially in 2011. Other acceleration features are also planned for Power8, IBM said.
The deal with SAP underscores IBM's increasing focus on tailoring its server technology to in-memory databases. "Data is never stagnant, it's alive. You've got to be able to move it," Doug Balog, general manager of IBM Power Systems, stressed in a company video. "So we went down this path of more processors, more cores, more I/O bandwidth, more memory bandwidth, more caching." The result, Balog claimed, is the same-day delivery of query reports that used to take a month.
SAP announced plans in May to make its HANA "Business Warehouse" generally available on IBM Power8 servers. Support for the SAP HANA Business Suite on Power is scheduled to be available by the end of the year.
IBM said its Power platform for SAP HANA will be available in several configurations aimed at different data warehouse sizes, starting with a set of "start configurations" that can be tailored to customer requirements. IBM is also offering special pricing on memory and core configurations running SAP HANA workloads. (It added that other workloads would run in virtualized systems.)