Skip to main content

Adopting SCI and SSA interconnects in advanced microprocessor based PC-servers

Exploitable results

Enterprise servers are increasingly being created by connecting together off-the-shelf personal computer (PC) or workstation components within so-called clustered systems. For these, the technology used to link the PCs together (the interconnect) is critical in providing the performance and fault tolerance required. An open cluster solution has been developed using an interconnect based on Dolphin's peripheral component interconnect-scaleable coherent interface (PCI-SCI) adapter, SCI switch and driver software. It makes reliable, high bandwidth links available to user-level applications with very small message delay (latency). This significantly reduces the performance overhead introduced by the interconnect itself when passing messages between nodes (the most critical interconnect requirement for effective clustered systems) as well as for block transfers. The solution is open, with a portable base platform allowing easy integration to various applications, and is scaleable to meet the requirements of different systems.
DREAMS is an integrated solution that provides a reliable, consistent and continuos access to data information for mission critical applications. Built on top of industry-standard interconnected computer servers and off-the-shelf software platforms, DREAMS is an open multinode system easy to manage, highly reliable, flexible to grow and meet the requirements of both workgroup and enterprise information management systems. DREAMS demonstrates a cost effective multinode cluster solution. It addresses the needs of mission critical applications by providing continuous processing and data availability, thus minimizing the downtime costs. It offers a flexible scaleable computing platform, capable to expand as demand grows. It takes advantage of industry standard building blocks which compete at different levels to the global solution. Server platforms are interconnected to form a multinode system thus achieving performance and availability advantages. Scaleability is provided by the ability to integrate multi-processor servers and to expand the cluster configuration up to four nodes. Connection among nodes is realized by means of industry standard SCI link and a SSA channel. SCI makes a reliable high bandwidth and low latency link available to the clustered system, significantly reducing the overhead due to message passing among the nodes. The access to shared disks benefits from the high-speed redundant channel provided by SSA. Off-the-shelf software applications on the most common operating system platforms provide continuous processing and data high availability: should a server fail, other servers in the clustered system will act on behalf of the failed one. A stock-exchange application validates the multinode platform capabilities in a real environment, taking advantage of the features offered by the cluster solution in terms of availability and scaleability.