Network Flow Shaper for Incast Buffer Overflow

The patent segments the data flow by introducing intermediate buffer nodes between data sources and the receiving switch. Each intermediate node buffers and forwards packets individually, dividing the large parallel data stream into smaller manageable units that can be processed without overwhelming the receiving switch buffer.

The patent introduces intermediate buffer nodes as mediators in the data path. These intermediate nodes act as buffers that receive packets from multiple sources and forward them to the destination, preventing direct buffer overload at the receiving switch while maintaining parallel data transfer capability.

Instead of increasing the buffer capacity at a single receiving switch, the patent segments the buffering function across multiple intermediate nodes distributed in the network. Each intermediate node maintains a small buffer, collectively providing sufficient buffer capacity without requiring any single switch to have excessive buffer resources.

The patent moves the buffer capacity from a single dimension (one switch's buffer) to multiple dimensions (distributed across several intermediate nodes). This spatial distribution of buffering capacity allows the system to handle larger data volumes without requiring any single device to have proportionally larger buffer resources.

The patent implements self-service by having intermediate buffer nodes automatically buffer and forward packets based on local conditions without requiring complex centralized control. Each intermediate node independently manages its own buffer, reducing the need for sophisticated network-wide coordination while still achieving load distribution.

The patent applies preliminary action by having intermediate nodes buffer packets before they reach the receiving switch. This advance buffering prevents the receiving switch from being overwhelmed by sudden data arrivals, smoothing out traffic patterns proactively rather than reactively.