Clusters consist of multiple interconnected servers and storage devices forming what appears to users as a highly available system. A typical cluster consists of a front-end server handling all user interactions, a private network isolated from the outside for reliability and security reasons, and computing nodes on the private network each with one to two processors and 4-16GB of DRAM. Load balancing and high availability are key features of a cluster system. Standards-based Linux clusters are rapidly becoming mainstream due to the attractive price/performance advantage they offer over custom solutions. These systems feature high internal memory-to-processor communications rates and can handle mathematically complex problems. The key to delivering high computational throughput is the proper balance between processor speed, I/O throughput, memory bandwidth, and latency. To ensure high availability, designers implement error correction coding (ECC) and chipkill on the memory subsystem. In addition to the above features, clusters are scalable, making it relatively easy to increase their capability. Clusters may be designed by using servers in various form factors depending on the application. Netlist supports these applications with 1-4GB RDIMMs (registered DIMMs) with speeds up to PC3200.
|
