Cluster Design Services
Designing a cost effective high performance cluster that fits your needs requires more than just an in-depth knowledge about computer hardware, it requires industry-specific experience. Aspen has that expertise. We specialize in building small and medium HPC clusters, database clusters, storage clusters, visualization Clusters, and large HPC Clusters. While many of the basic design considerations are identical, each type presents specific challenges and has areas where more attention should be given.
Aspen provides several different levels of Cluster Design services, depending on your cluster size and type. Your design process begins the very first time you call us. Your sales engineer will follow a standardized design flow process to ensure that he or she collects and understands your specific requirements. This due diligence is summarized in our HPC Buyers Guide and your sales engineer may direct you to that page while they discuss your cluster design and implementation.
Cluster Design Process - Small, Medium, Database, Storage, and Visualization Clusters
- First, we learn your application mix. Knowing your applications performance characteristics is critical to ensuring that your cluster investment performs appropriately. Many different factors such as communications latency requirements, floating point versus integer performance requirements, local vs. remote vs. parallel storage, and memory footprint per node matter a great deal to the efficiency of your code. Special hardware requirements, environment integration, and other factors can directly affect the actual effectiveness of your solution, and all those must be taken into account to truly understand your solution parameters, Some commercial applications and widely used models have well-known requirements and configurations that are well documented, while others, including many in-house codes, are less well understood. That's where benchmarking comes into play.
- With our benchmark clusters, we can help you test-drive your software on several different architectures and interconnects, and determine what configuration options best fit your requirements and provide the best performance for your codes at the lowest price. Factors such as what combination of file systems, compilers, MPI implementations, and interconnects that you use can severely affect your performance, and we strive to allow you to easily test several different configurations so that you can quickly find your optimal solution. From bleeding edge to the best price leaders from AMD and Intel, we have them.
- Many HPC vendors do not offer benchmarking because it is very expensive. At Aspen, we do, and we expend quite a bit of capital and engineering resources to help you determine the best fit for your needs. Unlike many vendors who wish you to buy whatever standard package they currently offer because it is less work for them to build and deploy, we want the best solution for you. We may not have the exact speed processor you're after in our benchmark cluster today, or many nodes of the type you want to test at any given time, but we strive to give you at least a small number of many different types so that you can quickly see results on as many different architectures and configurations as possible. Even larger node scaling benchmarks can be arranged, but we usually need a bit of time and coordination to set those up. In many cases, we may have already ran your particular application or code, and can quickly steer you toward the architecture, interconnect, and software configuration that we have learned is best for your needs.
