The name of the game today in business is speed: faster deliveries to customers, quicker responses to customer questions, speedier pricing decisions. The ability to perform these tasks and others is what often separates a thriving and profitable business from an also-ran.
Real-time or near real-time decision making requires an incredible array of technology all working in concert to provide decision makers with the right data at the right time. These systems sift through massive amounts of data in order to find, process and deliver information in a usable form.
Traditional systems and database tools are often not up to the challenge of serving the needs of today’s fast paced businesses. Fujitsu recognized this years ago, leading to our development of Software on Chip features in the Fujitsu SPARC64 X/X+ processors. Software on Chip can radically increase the performance of Oracle database queries when using the Oracle Database In-Memory option.
Running the database in memory makes a huge difference in performance
Queries can run at bus speeds, with radically lower latencies, plus higher bandwidth, than even the fastest connections to disk arrays. With Oracle Database In-Memory, data is stored in both traditional row format (for OLTP queries) and the new in-memory column format (great for analytic queries) simultaneously – thus optimizing the database for both types of workloads and greatly reducing processing time.
With the Fujitsu M10 server, customers can configure up to 1TB of memory per physical processor socket – and as much as 64TB in a single system. This allows customers to run even their largest databases in memory. However, there is much more to this story. Fujitsu M10 Software on Chip technology supplies performance above and beyond what one gets from the in-memory database technology alone.
One of Software on Chip’s most significant features is the implementation of SIMD (Single Instruction Multiple Data) vector processing technology. Let’s take a closer look at this technology and what it means to system performance:
In non-SIMD systems, such as older SPARC-based and other platforms, only one data entry can be loaded into each register, meaning only one comparison can be executed at a time. The processor could execute two instructions with each cycle, so you would get two data entries evaluated with each cycle.
By contrast, using Fujitsu M10 SIMD and Oracle Database In-Memory, data is stored contiguously in fixed length chunks. With the first implementation of SIMD in SPARC64 X processors, up to 8 data entries can be processed per cycle. SIMD capabilities were increased in the SPARC64 X+ processor so that now up to 16 data entries can be processed per instruction, and two of these instructions can be processed simultaneously. That means that a total of 32 values are evaluated with each cycle – a huge improvement!
There are some caveats here. The first is that you need a Fujitsu M10, and the latest SPARC64 X+ based M10 systems to provide the greatest Software on Chip SIMD benefits. Also, to get the full measure of performance available from the Fujitsu M10 system, there are some Solaris and Oracle Database pre-requisites:
- For SPARC64 X based M10 systems, you need to run at least Solaris version 11.1 and Oracle Database 12cR1 Enterprise Edition or later with the In-Memory option.
- For SPARC64 X+ based M10 systems, you need to run at least Solaris version 11.2, Oracle Database 12cR1 Enterprise Edition or later with the In-Memory option, and Oracle Database patch 20342458 (or Oracle Database Proactive Bundle Patch 188.8.131.52.7 or later).
So how big of a performance boost are we talking about? Potential benefits are substantial. Here are some impressive results from our verification testing using the Star Schema Benchmark (SSB):
- Oracle Database In-Memory is 16 to 18 times faster than when the In-Memory option is not used.
- When both Oracle In-Memory and Fujitsu SIMD are used, performance improves by 49 to 190 times compared to when both are disabled.
- When Oracle Database patch 20342458 is applied (enabling the “full power” of SPARC64 X+ processors), and when Oracle In-Memory and Fujitsu SIMD are used, processing time is improved by 73 to a whopping 278 times.
Please keep in mind that data and database queries vary widely. Exact performance benefits are tied to data entry size and the types of analytic queries run. As you would expect with technologies as complex as Software on Chip SIMD and the vast array of possible database workloads, we are not saying that every single query is going to be exactly 278 times faster. Some may not quite reach that level; some may exceed it. But, that should not diminish the fact that Oracle Database In-Memory and the Software on Chip technologies baked into each Fujitsu M10 processor can play a huge role in delivering those all-important real-time decisions.