Storage Appliance Compute Offload for FaaS Bandwidth Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In data centers, processing of compute-intensive and data-intensive functions in Function as a Service (FaaS) models is inefficient due to heavy network and memory bandwidth usage, leading to slowed application throughput and increased costs, especially when using field programmable gate arrays (FPGAs) that require data transfer between compute servers and storage devices.
Innovation Solution
Implementing a data center architecture that offloads compute tasks from compute servers to storage appliances, allowing for the execution of functions closer to mass storage devices, thereby reducing data traffic and improving performance through the use of storage-centric compute offloads and flexible software and hardware accelerators.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If functions are executed on compute server CPUs with data read and processed during execution, then processing capability is provided, but network and memory bandwidth are heavily consumed, becoming bottlenecks that slow down application throughput
Solution Approach 1:
The patent extracts the compute-intensive function execution from the compute server CPU and relocates it to a storage appliance. This separation removes the processing bottleneck from the network and memory subsystems, allowing data to be processed at the storage appliance without requiring continuous high-bandwidth communication between compute servers and storage devices.
Solution Approach 2:
The storage appliance acts as an intermediary between the compute server and the mass storage devices. It receives data from the compute server, executes the compute-intensive functions locally, and returns results to the compute server, thereby mediating the data flow and reducing the burden on network and memory bandwidth.
2Speed
If FPGAs are used to execute functions in compute servers, then processing speed is improved, but data flow control and result transfer still require heavy network and memory bandwidth usage
Solution Approach 1:
The patent extracts the FPGA-based function execution from the compute server and relocates it to the storage appliance. This maintains the high processing speed benefits of FPGAs while removing the associated network and memory bandwidth consumption from the compute server ecosystem.
Solution Approach 2:
The patent changes the spatial dimension of computation by moving processing resources from the compute server dimension to the storage appliance dimension. This dimensional shift allows processing to occur closer to the data source, reducing the need for data movement across the network and memory subsystems.
3Productivity
If compute-intensive and data-intensive functions are processed by data center compute servers, then function execution is provided, but costs increase due to per core usage or per function execution pricing
Solution Approach 1:
The storage appliance is designed to serve multiple functions: it acts as both a storage device and a compute engine. By consolidating these functions into a single system, the patent eliminates the need for separate compute server resources, thereby reducing costs associated with per-core usage or per-function execution pricing.
Solution Approach 2:
The patent merges storage and compute capabilities into a single storage appliance. This consolidation allows the same hardware resource to perform both data storage and compute-intensive function execution, reducing the total cost of ownership compared to using separate compute servers and storage devices.
Data Source
AI summary
Examples may include a storage appliance having a mass storage device and a compute engine communicating peer-to-peer with each other, with the compute engine including a programmable logic component to execute a function to read data from the at least one storage device, process the data; and write data to the at least one storage device.