GPU vs CPU Smackdown : The Rise of Throughput-Oriented Architectures

In some ways the original Amazon cloud, the one most of us still live in, was like that really cool house that when you stepped inside and saw the old green shag carpet in the living room, you knew the house hadn't been updated in a while. The network is a little slow, the processors are a bit dated, and virtualization made the house just feel smaller. It has been difficult to run high bandwidth or low latency workloads in the cloud. Bottlenecks everywhere. Not a big deal for most applications, but for many high performance applications (HPC) it was a killer.

In a typical house you might just do a remodel. Upgrade a few rooms. Swap out builder quality appliances with gleaming stainless steel monsters. But Amazon has a big lot, instead of remodeling they simply keep adding on entire new wings, kind of like the Winchester Mystery House of computing.

The first new wing added was a CPU based HPC system featuring blazingly fast Nehalem chips, virtualization replaced by a close to metal Hardware Virtual Machine (HVM) architecture, and the network is a monster 10 gigabits with the ability to specify placement groups to carve out a low-latency, high bandwidth cluster. Bottlenecks removed. Most people still probably don't even know this part of the house exists.

The newest addition is a beauty, it's a graphics processing unit (GPU) cluster as described by Werner Vogels in Expanding the Cloud - Adding the Incredible Power of the Amazon EC2 Cluster GPU Instances . It's completely modern and contemporary. The shag carpet is out. In are Nvidia M2050 GPU based clusters which make short work of applications in the sciences, finance, oil & gas, movie studios and graphics.

Click to read more ...

Implementing the Lustre File System with Sun Storage: High Performance Storage for High Performance Computing

Much of the focus of high performance computing (HPC) has centered on CPU performance. However, as computing requirements grow, HPC clusters are demanding higher rates of aggregate data throughput. Today's clusters feature larger numbers of nodes with increased compute speeds. The higher clock rates and operations per clock cycle create increased demand for local data on each node. In addition, InfiniBand and other high-speed, low-latency interconnects increase the data throughput available to each node. Traditional shared file systems such as NFS have not been able to scale to meet this growing demand for data throughput on HPC clusters. Scalable cluster file systems that can provide parallel data access to hundreds of nodes and petabytes of storage are needed to provide the high data throughput required by large HPC applications, including manufacturing, electronic design, and research. This paper describes an implementation of the Sun Lustre file system as a scalable storage cluster using Sun Fire servers, high-speed/low-latency InfiniBand interconnects, and additional networking and storage devices. Furthermore, this paper explores the use of the Sun Lustre file system at a shared government and education research site, including configuration information and details on testing that was performed on-site to evaluate the performance of Sun's scalable storage solution.

Click to read more ...

Sun Storage and Archive Solution for HPC

When designing data storage solutions for High Performance Computing (HPC) environments, IT architects strive to balance complex and often conflicting requirements. The need to manage a skyrocketing amount of data, along with the goals of controlling cost and immediate data availability, can make it difficult to meet HPC application demands within the constraints of today's IT budgets. To help customers address an almost bewildering set of architectural challenges, Sun has developed the Sun Storage and Archive Solution for HPC, a reference architecture that can be easily customized to meet specific application goals and business requirements. This article is intended for IT managers and storage architects familiar with HPC applications and data requirements in the organization. It assumes that the audience has a technical background and some familiarity with issues surrounding the task of configuring systems and storage.

Click to read more ...