Throttling Bandwidth Imbalanced Data Transfers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-performance network fabrics often exceed the write bandwidth of non-volatile memory (NVM) in high-performance computing systems, leading to network congestion due to bandwidth imbalances, which can degrade overall system performance.
Innovation Solution
A receiver-driven throttling transfer protocol that splits large message payloads into multiple chunks, scheduling their transfer to avoid overwhelming the receiver's memory bandwidth and network fabric, using techniques such as remote memory get operations and triggered operations to manage chunk get operations based on NVM write bandwidth and network capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If high-performance network fabric is used to achieve fast data transfer, then network bandwidth is improved, but the receiver's NVM write bandwidth becomes the limiting factor causing network congestion
Solution Approach 1:
The patent divides a large data transfer operation into multiple smaller chunk transfers. Each chunk is transferred individually with its own control message, allowing the receiver to process and acknowledge each chunk separately. This segmentation prevents overwhelming the NVM write bandwidth while maintaining high network utilization, resolving the contradiction between fast network transfer and NVM write capacity.
2Productivity
If conventional communication protocols are used to inject data as fast as possible, then network bandwidth utilization is improved, but the receiver is overwhelmed due to source/sink bandwidth mismatch
Solution Approach 1:
The patent implements a feedback mechanism where the receiver sends acknowledgments after successfully writing each data chunk to NVM. The sender uses these acknowledgments to regulate the timing and rate of subsequent chunk transmissions. This feedback loop ensures that data injection rate adapts to the receiver's actual write capacity, preventing buffer overflow while maintaining high productivity.
Solution Approach 2:
The transfer protocol dynamically adjusts the timing and pacing of data chunk transmissions based on receiver acknowledgments and NVM write status. Rather than using a fixed high-speed injection rate, the system dynamically modulates the data flow to match the receiver's instantaneous write capacity, resolving the contradiction between high productivity and reliability.
3Loss of time
If large payload transfers are performed in single operations, then transfer efficiency is improved, but network buffers become congested due to bandwidth imbalance
Solution Approach 1:
The patent segments large payload transfers into smaller chunk operations, each with associated control messages. This segmentation allows network buffers to drain between chunks rather than being continuously filled by a large monolithic transfer, reducing buffer congestion while maintaining acceptable transfer latency through pipelined chunk transmission.
Data Source
AI summary
Techniques are disclosed to throttle bandwidth imbalanced data transfers. In some examples, an example computer-implemented method may include splitting a payload of a data transfer operation over a network fabric into multiple chunk get operations, starting the execution of a threshold number of the chunk get operations, and scheduling the remaining chunk get operations for subsequent execution. The method may also include executing a scheduled chunk get operation in response determining a completion of an executing chunk get operation. In some embodiments, the chunk get operations may be implemented as triggered operations.