ScaleMP - Comprehensive solutions for HPC, scaling computing power and memory augmentation.
vSMP ClusterONE
Simplify your first HPC cluster and run one system.
The vSMP ClusterONE package simplifies cluster management by treating it as a single administrative task of transforming a cluster of physically separated servers, into one single logical computer, resembling a super performance workstation. Managing vSMP ClusterONE is as easy as managing a Linux workstation.
Immediately after deployment, users can interactively work with the system as they would with a workstation. The system runs standard Linux-type operating system management tools. The Linux volume management function creates a high-performance file system for temporary workspace (scratch). It is not necessary to manage networks or subnets and multiple operating systems.
The vSMP ClusterONE solution offers enhanced capabilities compared to standard cluster systems:
- Ability to work in batch and interactive modes.
- Single operating system - there is no need to provide the same software on multiple nodes.
- Built-in, shared input/output (I/O) operations, enabling local high-performance volumes of temporary workspace.
- The ability to use a high-performance InfiniBand network without wasting time learning and without the need for any drivers.
Until now, the transition from workstation-class systems to HPC processing environments has been a very complicated undertaking, requiring the use of expensive, specially trained professionals, even for small cluster deployments. With vSMP ClusterONE, there are no barriers. Regardless of the number of server nodes, vSMP ClusterONE behaves like a large workstation, with a large number of cores, a large amount of RAM and a huge number of I/O operations.
Brief specification:
- Up to 32 nodes with dual processors, up to 6GB/core;
- Recommended local storage on nodes (for high-speed working storage area with software - RAID on all disks);
- Redhat, CentOS, or SUSE operating systems;
- The vSMP Productivity Pack for seamless operation and automatic tuning.
vSMP ServerONE
Scale the computing power or memory of the server.
The vSMP ServerONE solution combines multiple standard x86 servers into a single virtual, high-performance system, an excellent alternative to the expensive multiprocessing solutions (SMP or NUMA) used until now. Using the vSMP ServerONE software suite, computing power and memory are not tied together, as they were before, and users can scale selected system attributes.
vSMP ServerONE transforms multiple servers, into a single computer system, seen by the operating system, applications, administrators, designers and users, as a single entity running only one copy of the operating system. Each processor in the combined system has access to all memory, allowing applications to scale, using OpenMP-type models of parallel thread execution.
The functionality of vSMP ServerONE can be obtained on demand, transforming any collection of nodes connected to the same network, into one single system. Thus providing a true software-definable server infrastructure for high performance computing (HPC) and enterprise-class data centers.
Main features:
- Running any HPC application delivering the highest performance for SMP (symmetric multiprocessing) applications.
- Cluster capabilities without the need for dedicated hardware and components.
- On-demand SMP processing from a single management point, increasing equipment utilization and reducing costs.
- Using the latest generation of processors and connectivity interfaces to deliver superior performance at a reasonable price.
- The ability to selectively scale parameters to allow the system to match the anticipated load and incur the cost of only what is used.
The maximum number of nodes making up a scalable SMP solution is 128 with up to 32768 processors and 2 PB (petabytes) of RAM.
vSMP MemoryONE
Increase memory size and reduce DRAM costs.
The vSMP MemoryONE solution enables the use of non-volatile content memory to augment or replace DRAM. Many companies and institutions are using direct in-memory processing to achieve very high application performance and get significant results using advanced business analytics. However, direct in-memory processing can be limited by a number of requirements related to the available amount of memory per server. This is due to the architecture of today's servers, as well as the prohibitive price of DRAM.
vSMP MemoryONE extends the RAM of standard x86 systems using NVMe SSDs. As a result, it enables much more cost-effective use of larger RAM while maintaining DRAM-like performance.
The solution allows:
- Obtaining a very large working memory:
For direct in-memory processing such as in-memory databases, analytics and numerical simulations.
- Achieve a lower total cost of ownership (TCO):
Increase system memory using NVMe instead of DRAM, allowing server/system consolidation by increasing the memory/processor ratio.
- Easy to implement and use:
No modifications to the operating system or applications are required.
Main functionalities:
- Scaling operating memory to tens of terabytes;
- 8 times or more the amount of memory compared to DRAM;
- Performance similar to DRAM;
- Up to 70% savings;
- Total transparency for Linux operating systems and applications;
- Enabling server consolidation;
- Choice of hardware: support for Intel and AMD processors, NVMe SSDs from multiple vendors, certification for all leading server manufacturers;
- Also available on demand in the Amazon AWS cloud;
- Support for major Linux operating system distributions.