Behind the scenes @ Labs
cancel
Showing results for 
Search instead for 
Did you mean: 

How HPE Superdome Flex’s in-memory computing gives you a head start on Memory-Driven Computing

ssinghal

header.jpg

In a previous blog, I described some of the research results from The Machine project at Hewlett Packard Labs. I commented in that blog that most of the measured results were obtained on an HPE Superdome X system, and that software developers do not have to wait for another three years to get the promising performance improvements we demonstrated.

Yesterday, Hewlett Packard Enterprise introduced the HPE Superdome Flex, which combines technology from both the Superdome X and the MC990X (previously the SGI UV300) system. It brings together many of the mission-critical features of the Superdome X and the scalability of the MC990X to create a platform that can scale from a 4-socket, 6 TB memory enclosure all the way to a 32-socket, 48 TB system and can scale effectively as workloads increase.

What may not be obvious in these announcements is that the Superdome Flex is also an ideal platform for Memory-Driven Computing.

It brings us one step closer to realizing our vision for The Machine. Because Superdome Flex has been designed with Memory-Driven Computing principles, it offers some unique advantages to address data-intensive workloads. At Labs, we have been working closely with our Mission-Critical R&D teams to develop a shared vision that gets us even closer to true Memory-Driven Computing.

Let’s dive deeper into the technology

Enclosures within the Superdome Flex are connected using a high-performance all-to-all memory fabric that is used to access memory across the enclosures by the operating system. Normally, the system runs a single instance of an operating system, and behaves as a single scale-up machine.

However, it is also possible to run multiple instances of operating systems on each enclosure by partitioning the system. A number of enclosures can be combined into partitions, and each partition can run its own operating system. What I find most interesting is that the partitions can also share memory across the operating systems using the memory fabric. Each enclosure can contribute memory to a global pool, and it is possible to present this global pool to applications running in the different operating systems as a common shared resource.

Using the Superdome Flex as a partitioned systemUsing the Superdome Flex as a partitioned system

The image on the left above shows a “normal” configuration with a single system image. The image on the right shows a partitioned system, where each partition has contributed some memory to create a virtual global memory pool that is shared across the operating systems. The system on the right now has many of the same properties we’ve described as at the core of our Memory-Driven Computing architecture: A shared global memory pool, a high performance memory semantic fabric, and a horizontally scalable compute environment.

Memory Driven Computing PrinciplesMemory Driven Computing Principles

While the memory in a Superdome Flex system is not physically separated out onto the fabric as in The Machine prototype, with small changes in the operating system, it is possible to make the application believe that it is running on The Machine (within the limits posed by 32 sockets and 48 TB memory).

This development provides our customers and partners with yet another opportunity to realize the benefits of Memory-Driven Computing right now, and is another example of how HPE is leading the industry with cutting-edge in-memory computing solutions.

For those of you attending HPE Discover 2017 in Madrid, register now to join me and HPE Distinguished Technologist Mike Woodacre – chief engineer for Superdome Flex – for a conversation about what this means for our customers, and how to get started.

0 Kudos
About the Author

ssinghal

Software and Applications for The Machine