Compaq ProLiant 1000 Virtual Interface (VI) Architecture - The New Open Standa - Page 4

Virtual Interface (VI) Architecture - The New Open Standard for Distributed Messaging Within a Cluster 4 VI Architecture Implementation Compaq has been working with Tandem Computers in developing a VI Architecture-based adapter by engineering VI Architecture specifications into the hardware adapter. This will provide the high-speed and low latency required ensuring superior performance in a cluster. Figure 1 illustrates how VI Architecture would be implemented within an industry standard-based SAN. Figure 1. VI Architecture Structure within a clustered System Area Network (SAN) System Area Network and VI Architecture High Speed Messaging Seamless Growth Scalable Performance Research Applications OSV Specific Application Specific Legacy Upper Level Protocols OS and HW independent HW Interface VIPL Virtual Interface OS Dependencies SW Interface VI Architecture for SAN OS and HW independent SAN/VI NIC OS and HW dependent VI Kernel Agent OS Kernel Kernel Client s System Area Network The roots of VI Architecture are in server messaging without allowing the operating system to relinquish memory to an application if other applications or operating system services need memory. In other words, VI Architecture will reserve memory so that distributed calls from other clustered nodes will not have to wait for memory to be mapped. One of the key benefits of VI Architecture is its ability to reduce or remove the protocol stack(s) for send/receive and read/write for applications, thus removing a large percentage of operating system and server processing overhead for functions such as LAN and WAN packetizing and depacketizing. The operating system will still continue to provide security and messaging setup for applications as well as less common distributed processes where kernal pass through and operating system services are used. The VI Architecture provides a widely-supported set of commands to send reliable distributed messages between clustered servers, allowing applications to use reserved or pinned memory directly for message services between clustered servers. The VI Architecture also specifies a set of pre-defined queues and allows these queues to be resident on the VI Architecture compatible Cluster Interconnect Adapters such as future Tandem ServerNet adapters and OEM adapters that ECG098/0998