Oracle SPARC

SPARC (Scalable Processor ARChitecture) is the scalable architecture of RISC microprocessors which is originally developed in 1985 by SUN Microsystems company. The architecture of SPARC is open.

2017

SPARC M8 release

On September 20, 2017 Oracle announced the next generation of SPARC M8 chips. According to developer company, SPARC M8 is twice faster than chips of Intel in online transaction processing (OLTP) and semikratno — in work with analytics of databases.

The company made focus in work on SPARC M8 on Silicon Secured Memory technology. It helps to prevent widespread cyber attacks - overflow and rewriting of the buffer. End-to-end enciphering is supported, there are building tools for the Data Analytics Accelerator (DAX) accelerator increasing performance and efficiency of Oracle Software in Silicon (software on a crystal). The processor consists of 32 cores (eight flows on a core) that allows it to process up to 256 flows at the same time^[1].

Oracle SPARC M8, (2017) processor

SPARC M8 is made on 20-nanometer technical process. To 32 cores of the processor the cache of L3 with an amount of 64 MB, a cache of data of L2 of 128 CB on a core and 256-KB cache of L2 commands separated into four cores is created. Separately 32-KB cache of L1 commands and a 16-KB cache of data of L1 are available to each core. Operating frequency of the processor ~ 5 GHz. M8 supports hardware acceleration of enciphering for formats: AES, Camellia, CRC32c, DES, 3DES, DH, DSA, ECC, MD5, RSA, SHA-1, SHA-3, SHA-224, SHA-256, SHA-384 and SHA-512.

The processor has the block of commands and executive blocks. To them treat the integer pipeline with extraordinary accomplishment and support of Quad Issue, the block for numbers from a floating comma and the integral cryptographic unit on a core. Thus operation of applications of databases is supported.

It is supposed, SPARC M8 will work under management OS Solaris at least till 2034.

Oracle carried out terrible reduction of SPARC and Solaris developers

Oracle carried out by summer of 2017 the large-scale staff reduction which affected generally developers of the Solaris operating system and architecture of SPARC microprocessors and also tape libraries and storages. Judging by records on a board of messages of The Layoff, about 2.5 thousand employees^[2] were in total dismissed^[3].

The company in this respect did not make any statements — dismissals were announced on the Internet by her employees who fell under reduction. In particular, about it wrote to social networks Twitter Drew Fisher, the senior developer Python and OpenStack for Solaris.

It's true. Oracle reduced most of Solaris (and others) today" — he reported. Fischer also lost work about what he wrote separately: "Oracle presented to Dobby a sock! Dobby is free! Somebody needs Python developer with experience 15+ years?"

2016

API для Data Analytics Accelerator

On March 15, 2016 the Oracle corporation announced release of the open API interface and an array of instrumentation of development for the Data Analytics Accelerator (DAX) accelerator in SPARC M7 processors.

These resources give to developers an opportunity to create platforms for data analysis. The initiative of Software in Silicon Developer Program helps developers to study examples of scenarios of use and a program code, to test and check possibilities of DAX on acceleration of analytical applications using Software in Silicon technology.

Representation of Data Analytics Accelerator is (DAX) of Deep Dive (2016)

High-performance data analysis is crucial for different key scenarios of use, such as analysis of attendance of the websites, opinions on social networks, behavior of buyers, etc. Participating in our Software in Silicon Developer Program program, developers can apply DAX technology to a broad spectrum before of unsolvable problems now in the analytical systems as we integrated acceleration of data analysis in processors, having provided the unprecedented scanning speed of data reaching 170 billion lines per second. John Fowler, executive vice president for the Oracle Systems direction

Oracle entered a number of functionality of the class Software in Silicon to the 32-core, 256-line SPARC M7 processor, having built in program functions of higher level design of the processor. One of the most impressive new opportunities implemented within innovations of Software in Silicon in SPARC M7 is DAX technology which provides unsurpassed efficiency of the analysis.

The DAX technology adds the means of processing allowing to perform functions, selecting data — such as Scan, Extract, Select and Translate. SPARC M7 DAX provides acceleration of these analytical primitives on the selected physical coprocessor separate of standard computing cores of the processor. Originally the DAX technology was implemented for Oracle Database 12c and all applications which work at it. It allowed to accelerate the analysis on all Oracle applications, independent suppliers (ISV) and applications of own development of clients. For March 15, 2016 the DAX technology is available to any applications.

Transactions of large-scale scanning and filtering become simpler thanks to transparent use of 32 selected DAX coprocessors in the SPARC microprocessor which operate with memory directly - at the speeds of the memory bus reaching 160 Gbps between a cache and DRAM. These accelerators for the first time implemented in the processor for ensuring the highest performance levels and efficiency, can be used by developers via API interfaces in Oracle Solaris 11 now and be applied in different scenarios of use.

As an example of integration of Data Analytics Accelerator into scenarios of machine learning and use of Big Data Oracle developers showed how DAX allows to accelerate considerably Apache Spark which became one of the most popular projects on processing of sets of Big Data. Within this project engineers used DAX with Apache Spark to filter one billion lines of data in RAM in 3D - a cube so quickly that interactive data analysis became possible.

Constructive opportunities of SPARC M7 and DAX

• capacity of RAM: thanks to capacity of memory in 160 Gbps, the SPARC M7 processor has an opportunity to quickly supply with data both DAX coprocessors, and main cores.
• movement of load of DAX: main cores are released for accomplishment of other tasks.
• effective decompression in combination with processing in RAM: the decompression in the DAX coprocessor is executed much quicker, than in program implementations. The decompression with scanning, is executed for one transaction, excludes unnecessary transactions of data transmission in RAM and back. Results from DAX come to a processor cache for increase in efficiency of computing processing.
• comparison on ranges (range scan) using DAX: many real analytical requests to databases are written for data retrieval between specific dates, on different price ranges of goods, etc. DAX executes comparison on ranges for one transaction, with the same speed, as separate comparisons. When using other processors extra computing time for each comparison is required.
• exception of "washing away" of a cache of the processor: DAX executes the majority of calculations without need to store intermediate data in a cache, releasing a processor cache for other tasks.

Open API interfaces are available to Oracle Data Analytics Accelerator free of charge through a cloud platform of Software in Silicon Cloud. Developers can join this community, for creation of analytical applications of new generation for work with Big Data.

SPARC S7

On June 30, 2016 the Oracle corporation provided SPARC S7 processors^[4].

Processors are created on SPARC M7 core, but instead of 32 computing cores - eight. It allowed to create servers on S7 platform more compact, energy efficient and available at the price.

Sparc S7 processor, (2016)

At the price of the standard standard server customers can have all opportunities of a corporate cloud with such properties as security, simplicity and efficiency. Effective performance on a core at the new systems of Oracle for 50%-100% exceeds indicators of standard servers h86, at accomplishment of Java applications servers on S7 processors are ahead of competitors by 1.7 times, and for databases this coefficient makes 1.6. As for analytical tasks, in it to them is not present equal at all – they are solved up to 10 times quicker thanks to the accelerator of transactions borrowed from M7. Existence of open interfaces of programming (API) makes these opportunities available and to third-party developers. And the built-in security aids allow customers to create completely protected cloud with end-to-end data encryption. Marshall Choy, the vice president for development of system products from Oracle headquarters

The line of systems on the basis of S7 processors will give to Oracle the chance to expand a range of offers, to provide to customers of a system for different tasks – from high-performance platforms based on M7 to servers of the middle class on S7 processors. The last will become an optimal solution for ERP systems, applications of a back office, cloud loadings and so forth.

2015: SPARC M7

On October 29, 2015 the Oracle company provided new generation of SPARC RISC processors and servers on its basis^[5].

SPARC M7 (2015)

Feature of SPARC M7 - close integration of the hardware and the software, in particular, implementation at the level of silicon of the processor of the software functions improving performance of databases and security.

In comparison with the previous generation of SPARC M6, in the SPARC M7 processor the number of cores increased from 12 to 32, for 15% the clock rate - increased up to 4.1 GHz, volume built in a cache memory is four times more and memory bandwidth is doubled. The Security in Silicon function accelerates data encryption and reduces load of main cores at accomplishment of this transaction, and Silicon Secured Memory improves data protection, being in RAM, using the special color bit key.

The chart of system performance on the SPARC M7 (2015) platform

The SQL in Silicon function is implemented using coprocessors to which all cores of SPARC M7 are supplied. These coprocessors undertake processing of problems of data compression in memories, storage scans and filterings, exempting from them cores.

By data Oracle, when using SQL in Silicon database requests are processed 10 times quicker. Support of this function is implemented in the version DBMS Oracle Database 12c with the option In-Memory.

Representation of SPARC M7 (2015)

SPARC M7 processors are used in SPARC systems of Oracle, including the supercomputer SuperCluster M7 complex and servers of the SPARC T7 series with number of processor sockets from 1 to 4 and the SPARC M7 series, scalable from 8 to 16 sockets.

Notes