Parallel Path Cache Pre-staging for Storage Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current storage systems face inefficiencies in pre-staging data into cache due to the time required for transmitting in-band cache hints and loading data into cache, which can take hundreds to thousands of microseconds, impacting performance and reliability.
Innovation Solution
The method involves utilizing parallel paths of differing performance by transmitting a first command over a faster path to prepare for and influence the execution of a second command over a slower path, optimizing data pre-staging and cache management by pre-populating the cache before the actual data request.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If in-band cache hints are transmitted over a conventional communication path, then the storage system can receive cache management instructions, but the transmission takes hundreds of microseconds causing delay
Solution Approach 1:
The patent segments the cache management process into two independent parts: (1) sending the cache hint command over the faster PCIe path, and (2) sending the actual data transfer command over the traditional storage path. This segmentation allows the hint to be transmitted independently and in advance, eliminating the sequential dependency that caused the delay.
Solution Approach 2:
The storage system performs preliminary action by receiving and processing the cache hint command before the actual data transfer command arrives. The hint is processed in advance to pre-stage data in the cache, so when the data transfer command arrives, the data is already ready, eliminating the need to wait for hint processing during the data transfer.
2Productivity
If data is loaded into cache before delivery to application, then cache can improve I/O performance, but loading data into cache takes hundreds to thousands of microseconds
Solution Approach 1:
The system performs preliminary data staging in the cache based on the hint command received in advance. By processing the hint and pre-loading data into cache before the application's actual data request arrives, the system ensures that when the data is needed, it is already resident in the fast cache memory, eliminating the need for slow disk access at the moment of demand.
Solution Approach 2:
The patent separates the cache management operation (triggered by hint) from the data transfer operation. This allows the cache to be populated independently in advance, and the data transfer to proceed separately, with both operations overlapping in time rather than executing sequentially, thus hiding the data loading latency.
3Device complexity
If a single communication path is used for both cache hints and data transfer, then system complexity is reduced, but efficiency is lost due to sequential processing
Solution Approach 1:
The patent divides the communication infrastructure into two separate paths: a fast path (PCIe) for control commands like cache hints, and a traditional path for data transfer operations. This segmentation allows each path to be optimized for its specific function and enables parallel processing of hint commands and data transfer commands, dramatically improving system throughput and efficiency.
Data Source
AI summary
A method for utilizing parallel paths of differing performance to improve efficiency is disclosed. In one embodiment, such a method includes transmitting, over a faster path, a first command to perform first actions intended to improve efficiency of second actions associated with a second command. The method transmits, over a slower path in parallel with the faster path, the second command. Alternatively, a method for utilizing parallel paths of differing performance to improve efficiency includes receiving, over a faster path, a first command to perform first actions intended to improve efficiency of second actions associated with a second command. The method executes the first command to perform the first actions. The method receives, over a slower path in parallel with the faster path, the second command and executes the second command to perform the second actions. Corresponding systems and computer program products are also disclosed.