PCIe Fabric Buffer for Virtual Memory Sharing

The patent introduces a buffer device as an intermediary between information processing devices to manage memory space. The buffer device receives write requests from one device and forwards them to the target device, eliminating the need for individual memory management at each device. This mediator approach reduces processing overhead while maintaining memory sharing capability.

The patent extends the address space by introducing a buffer device that handles address translation and forwarding. Instead of being constrained by the 64-bit PCIe address limit, the system uses the buffer device to manage addresses beyond this limit, effectively adding another dimension to the addressable memory space while maintaining direct access capabilities.

The patent creates a universal buffer device that can handle both cluster and non-cluster configurations. The buffer device receives packets from any information processing device connected via fabric, performs address translation, and forwards them appropriately. This multi-functional approach maintains high data transfer efficiency while enabling connectivity across different device types and configurations.

The patent extracts the page-based access restriction mechanism from the memory sharing system. By using a buffer device to manage address translation and forwarding, the system removes the limitation that prevents multiple devices from accessing the same memory pages simultaneously. This extraction of the restrictive mechanism enables true shared memory access while maintaining virtual memory capabilities.

The patent introduces dynamic address translation through the buffer device. Instead of using fixed memory addresses that require identical device configurations, the buffer device dynamically translates addresses based on the actual device identities and current system state. This dynamic approach maintains ease of memory access while enabling flexible device configurations and arbitrary device connections.