Software-defined composable server infrastructure: Pinpoint the exact computing power needed for in-memory databases and analytics with a single-system server that has up to 32,768 CPUs and 2PB memory, and more than one PFLOPs peak for technical computing application of a large-memory system.
vSMP ServerONE aggregates multiple, industry-standard, x86 servers into one single virtual high-end system. Finally, organizations have an alternative to expensive legacy multiprocessing (SMP or NUMA) that can easily scale for different system attributes. With vSMP ServerONE, computing and memory are no longer linearly tied.
With vSMP ServerONE, multiple servers appear to the operating system, applications, administrators, developers or users as a single entity, running only one copy of the operating system. Each CPU in the aggregated system has access to all of the memory, enabling applications to scale using thread-parallel execution such as OpenMP.
vSMP ServerONE functionality can be delivered on-demand, turning any collection of nodes connected to the same fabric into a single system, and providing a true software-defined composable server infrastructure.
ABOUT THE PRODUCT
vSMP ServerONE provides access to all the aggregated memory of multiple systems from single or multiple threads. By combining a number of servers together in order to expand the amount of directly addressable memory, developers and users can easily access many terabytes of memory without having to code for complicated distributed programming environments. vSMP ServerONE gives users a cost-effective system that scales the memory to the desired amounts.
vSMP ServerONE also provides enhanced capabilities over the traditional SMP systems by offering:
- Software-defined composable server infrastructure, which provides greater utilization by reusing the same servers’ hardware for both distributed and shared-memory workloads.
- Hard partitioning, which enables different groups to work on a dedicated resource rather than share the same system and interfere with one another.
- Parallel InfiniBand performance, which uses multiple HCAs in parallel to improve performance for the most demanding applications.
Further, vSMP ServerONE adds support for very large SMPs, which delivers greater system scalability – the largest single-systems available from any vendor.
As workloads change, a data center with vSMP ServerONE installed can react to these changing workloads. Large, virtual SMPs can be created as needed for users that need access to hundreds of cores and/or terabytes of memory. A virtual SMP can be provisioned on the fly by a number of popular cluster management applications and workload managers. When the challenging workload decreases or changes to smaller resource-intensive applications, the servers can be re-provisioned as standard nodes in the cluster to continue to do useful work.
The same principle applies for “fat nodes”: Customers no longer need to provision dedicated specialized systems with more memory. Rather, vSMP ServerONE can be used to aggregate the memory of several of the cluster nodes to serve a specific large-memory job, then be re-provisioned as stand-alone servers for distributed cluster jobs.
On-Demand Scale-Up: Flexible and Cost-Effective
Depending on workload requirements, all of the resources (CPUs, memory, and I/O) can be aggregated, or just the memory of the nodes. We call this “On-Demand Scale-Up,” as the organization can adjust the system exactly to the application needs. Read more about On-Demand Scale-Up here .
PRODUCT QUICK SPECIFICATION
- See www.scalemp.com/spec for certified hardware and operating systems
- Up to 32768 CPUs
- Up to 2PB of shared memory