Subscribe

Share

Interview

SPARC M7 Delivers

Software in Silicon performance and security are hallmarks of new Oracle processor.

By Tom Haunert

January/February 2016

Hardware and software are different. Hardware, starting with the processor, typically runs software that’s developed separately and usually by other companies. Organizations build software in reaction to hardware capabilities, and that reaction typically takes significant time.

Oracle Magazine sat down with Marshall Choy, vice president of systems product management at Oracle, to talk about Oracle’s new SPARC M7 processor and how it bridges the hardware-software divide, saves customers time, and delivers superior security.

Oracle Magazine: What are the key hardware improvements in SPARC M7?

Choy: Relative to the previous-generation SPARC T5 processor, we’ve doubled the number of cores in SPARC M7. SPARC M7 is a 32-core processor running at 4.15 GHz—a 15 percent higher core frequency over SPARC T5. We’ve quadrupled the amount of cache per core, and we’ve lowered the cache latency by about 30 percent. In addition, we’ve enhanced I/O, tripling I/O bandwidth and quadrupling the number of PCI Express links.

The end result of these enhancements for customers is two to three times more throughput performance as well as about 40 percent more single-thread performance coming out of each of those 32 cores on the SPARC M7 processor.

Oracle Magazine: In traditional computing, the processor is hardware that runs other software. SPARC M7 introduces the concept of Software in Silicon—what is that, and why is it important?

Choy: We saw the advent of 64-bit computing about 20 years ago. It delivered larger memory support and greater precision for computational workloads. But the hardware came before the software, and there was some lag between 64-bit hardware and 64-bit software availability.

About 10 years ago we saw the beginning of multicore and multithreaded processors. Sun Microsystems [acquired by Oracle in 2010] was at the forefront of that with the SPARC T1 processor. Multiple cores and multithreaded operations supported heavily virtualized environments and the beginnings of cloud computing.

We are giving our customers the ability to run programs safely in silicon-secured memory and to store and access information in a fully encrypted data center.

Now Oracle is looking at customer operations and how to improve business processes. We’re looking at existing software operations, including the secure processing and management of big sets of data, and how we can optimize these critical operations in silicon.

The difference between SPARC M7 Software in Silicon compared to 64-bit and multicore computing is we are delivering the hardware and software capabilities together, at the same time, so that customers can immediately adopt these capabilities.

Oracle Magazine: What are the Software in Silicon analytics capabilities of SPARC M7?

Choy: A lot of database development in the past was really chasing the capabilities available in the hardware. The difference now with SPARC M7 Software in Silicon and Oracle Database 12c is that these two technologies were developed side by side, by development teams working together.

We had SPARC M7 beta customers running the Oracle Database In-Memory option of Oracle Database 12c, and one beta customer was able to achieve 83 times faster analytics with Oracle Database In-Memory and SPARC M7 Software in Silicon.

SPARC M7 includes query accelerator engines to increase the efficiency and performance of data analytics. There are 32 data analytics accelerator engines on the microprocessor that accelerate database scan, range scan, bloom filter, and join operations with Oracle Database 12c.

Oracle Magazine: What are the Software in Silicon capacity capabilities of SPARC M7?

Choy: The capacity-in-silicon capabilities of SPARC M7 are very tightly coupled to the data analytics acceleration. With capacity in silicon, our lab testing has shown you can run a 1-terabyte database in 120 GB of memory, because of inline decompression. The combination of inline decompression and data analytics done in line at memory speed delivers huge performance gains and performance efficiency by reducing several steps to one and increasing analytics performance up to 10 times.

Oracle Magazine: What are the specific Software in Silicon security capabilities of SPARC M7?

Choy: Security was a big area of focus for SPARC M7 development, and specifically, there are two things we’ve delivered under the umbrella of Security in Silicon. The first is silicon-secured memory, and the second is cryptographic acceleration.

With silicon-secured memory, at the time memory is allocated to an application, that memory is locked so that only the owner can access it. If a malicious program tries to access the locked memory of a legitimate program, the malicious program is aborted. The buffer overread and overwrite types of attacks of the notorious Heartbleed and Venom malware applications are not possible against SPARC M7 silicon-secured memory.

Silicon-secured memory is always on—you can’t turn it off, and it’s the first hardware-based memory protection that will protect against both memory intrusion and erroneously written code.

The second area of focus for security enhancements in the SPARC M7 processor is around crypto accelerators. Encryption is nothing new; we’ve had software-based encryption for a long time, and we’ve had SPARC processor-based encryption since 2005. What’s new and different in SPARC M7 encryption is we’ve got 32 crypto accelerators per processor that contain a broad range of ciphers, including SHA-512 and AES-256. The encryption overhead is minimal—so minimal that customers can simply run everything fully encrypted with near-zero performance overhead.

With SPARC M7 silicon-secured memory and cryptographic acceleration, we are giving our customers the ability to run programs safely in silicon-secured memory and to store and access information in a fully encrypted data center.

Next Steps

READ more about
 Sparc M7.
 Software in Silicon.

 

Photography by Dmitry Ratushny,Unsplash