SCIENCE
3 Questions: Shai Fultheim on vSMP Foundation
"3 Questions" is a new series from SC Online that gives members of the community the opportunity to sound off on current events in their field of expertise. In this installment, Shai Fultheim, founder and CTO of ScaleMP, discusses his firm's vSMP Foundation solution for greater flexibility, lower costs, reduced complexity and the ability to better manage customer service-level expectations.
SC Online: What is server virtualization for aggregation, how is it different from traditional server virtualization and what are its implications for customers in the supercomputing space?
Fultheim: With both types of server virtualization the hypervisor enables the disaggregation of applications (the entire operating system, application and data stack) from the underlying hardware. Traditional server virtualization partitions a single system into multiple virtual systems to run multiple parallel workloads and provide much higher server optimization and a much higher return on investment rate. While this approach works for highly parallelized workloads or applications that have limited RAM requirements, it does not work well for those applications that gain performance benefits from multiple CPUs or require very large memory and which have traditionally been run on proprietary and specialized computing hardware.
Server virtualization for aggregation combines multiple x86 servers and their CPUs and memory to create a large virtual machine so that IT can run compute intensive applications that need so much memory and CPU power that organizations have previously only been able to run them on dedicated symmetric multiprocessing (SMP) systems. What this means for organizations with SMP needs is that they can use their commodity servers to create a virtual SMP with equal or better performance and at a much lower cost than if they were to go out and purchase a proprietary system.
SC Online: What are the key challenges you’re seeing customers facing in terms of supercomputing/symmetric multiprocessing systems, and are these hardware or software issues?
Fultheim: One of the biggest challenges we’ve seen customers face is trying to find a system that meets their hardware specifications and that’s not too expensive. To reduce costs, many organizations try to rewrite their code to work on a smaller system, or rewriting OpenMP code as an MPI application. That’s not something organizations should have to worry about, especially since tailoring code to a specific hardware architecture that may become outdated within a few years is a waste of time and money. With ScaleMP’s vSMP Foundation, organizations can have the flexibility to create a virtual system that meets the exact CPU and memory needs of their workload.
SC Online: ScaleMP solutions are a part of the San Diego Supercomputer Center. What are you doing there and what role do you see ScaleMP technologies playing in other industries?
Fultheim: ScaleMP is an integral part of the next-generation high-performance computer, named Gordon, a project at the San Diego Supercomputer Center (SDSC) at the University of California, San Diego. One of the key features of Gordon will be 32 "supernodes" based on new Intel processors available in 2011. Each supernode consists of 32 compute nodes, capable of 240 gigaflops/node and 64 gigabytes (GB) of DRAM. A supernode also incorporates 2 I/O nodes, each with 4 TB of flash memory. When tied together using ScaleMP's vSMP Foundation technology, each of the system's 32 supernodes has the potential of 7.7 TF of compute power and 10 TB of memory (2 TB of DRAM and 8 TB of flash memory). Once fully installed and configured, Gordon is expected to rate among the top 30 supercomputers in the world.
ScaleMP’s vSMP Foundation is used by high-performance customers running financial analytics, data warehousing, genomics, fluid dynamics, visualization, simulation, electronic design, data analysis and other large memory workloads across virtually every industry: energy, finance, government, manufacturing, healthcare and pharmaceuticals, technology and education. vSMP Foundation for Cloud can be used by service providers to provision a virtual SMP system in cloud environments, on the fly. This enables service providers to allocate systems that exactly meet customer specifications.
SC Online: What is server virtualization for aggregation, how is it different from traditional server virtualization and what are its implications for customers in the supercomputing space?
Fultheim: With both types of server virtualization the hypervisor enables the disaggregation of applications (the entire operating system, application and data stack) from the underlying hardware. Traditional server virtualization partitions a single system into multiple virtual systems to run multiple parallel workloads and provide much higher server optimization and a much higher return on investment rate. While this approach works for highly parallelized workloads or applications that have limited RAM requirements, it does not work well for those applications that gain performance benefits from multiple CPUs or require very large memory and which have traditionally been run on proprietary and specialized computing hardware.
Server virtualization for aggregation combines multiple x86 servers and their CPUs and memory to create a large virtual machine so that IT can run compute intensive applications that need so much memory and CPU power that organizations have previously only been able to run them on dedicated symmetric multiprocessing (SMP) systems. What this means for organizations with SMP needs is that they can use their commodity servers to create a virtual SMP with equal or better performance and at a much lower cost than if they were to go out and purchase a proprietary system.
SC Online: What are the key challenges you’re seeing customers facing in terms of supercomputing/symmetric multiprocessing systems, and are these hardware or software issues?
Fultheim: One of the biggest challenges we’ve seen customers face is trying to find a system that meets their hardware specifications and that’s not too expensive. To reduce costs, many organizations try to rewrite their code to work on a smaller system, or rewriting OpenMP code as an MPI application. That’s not something organizations should have to worry about, especially since tailoring code to a specific hardware architecture that may become outdated within a few years is a waste of time and money. With ScaleMP’s vSMP Foundation, organizations can have the flexibility to create a virtual system that meets the exact CPU and memory needs of their workload.
SC Online: ScaleMP solutions are a part of the San Diego Supercomputer Center. What are you doing there and what role do you see ScaleMP technologies playing in other industries?
Fultheim: ScaleMP is an integral part of the next-generation high-performance computer, named Gordon, a project at the San Diego Supercomputer Center (SDSC) at the University of California, San Diego. One of the key features of Gordon will be 32 "supernodes" based on new Intel processors available in 2011. Each supernode consists of 32 compute nodes, capable of 240 gigaflops/node and 64 gigabytes (GB) of DRAM. A supernode also incorporates 2 I/O nodes, each with 4 TB of flash memory. When tied together using ScaleMP's vSMP Foundation technology, each of the system's 32 supernodes has the potential of 7.7 TF of compute power and 10 TB of memory (2 TB of DRAM and 8 TB of flash memory). Once fully installed and configured, Gordon is expected to rate among the top 30 supercomputers in the world.
ScaleMP’s vSMP Foundation is used by high-performance customers running financial analytics, data warehousing, genomics, fluid dynamics, visualization, simulation, electronic design, data analysis and other large memory workloads across virtually every industry: energy, finance, government, manufacturing, healthcare and pharmaceuticals, technology and education. vSMP Foundation for Cloud can be used by service providers to provision a virtual SMP system in cloud environments, on the fly. This enables service providers to allocate systems that exactly meet customer specifications.
