Write-back Cache Replication via NIC Packet Transmission

Solution Overview

Problem

Write-back flash caching in data center environments faces performance issues due to high latency and potential data corruption or loss when replicating data between host systems, especially when using RAID mirroring schemes, which can result in performance that is no better or even worse than traditional write-through caching with enterprise-class storage.

Innovation Solution

Implementing a technique where a primary host system intercepts VM write requests, saves data in a local cache, and transmits data packets to a secondary host system via a network interface controller, acknowledging the write request completion once the data packet is transmitted, thereby minimizing performance loss and allowing the VM to continue processing without waiting for network roundtrips.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If RAID mirroring is implemented for write-back flash cache replication, then data reliability is improved, but write latency increases significantly due to sequential operations and network roundtrips

Engineering Contradiction:

Improvedata reliabilityVSAvoidwrite latency

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent applies preliminary action by having the primary host system transmit data packets to the secondary host system in advance, before the VM completes its write operation. The secondary system prepares to receive and store these packets proactively. This allows the acknowledgment to be sent back to the VM immediately after transmission begins, rather than waiting for the entire replication process to complete, thereby reducing write latency while maintaining data reliability through preliminary replication setup.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary mechanism where the primary host system acts as a mediator between the VM and the secondary host system. Instead of the VM directly interacting with the secondary system or waiting for full replication confirmation, the primary system manages the transmission and acknowledgment process. This intermediary role allows the VM to receive acknowledgment quickly while the background replication process continues independently, resolving the contradiction between reliability and latency.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If synchronous save to shared storage is performed, then data consistency is ensured, but VM write performance deteriorates due to additional I/O latency

Engineering Contradiction:

Improvedata consistencyVSAvoidVM write performance

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent applies preliminary action by having the primary host system transmit data packets to the secondary host system in advance, before the VM completes its write operation. The secondary system prepares to receive and store these packets proactively. This allows the acknowledgment to be sent back to the VM immediately after transmission begins, rather than waiting for the entire replication process to complete, thereby reducing write latency while maintaining data reliability through preliminary replication setup.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary mechanism where the primary host system acts as a mediator between the VM and the secondary host system. Instead of the VM directly interacting with the secondary system or waiting for full replication confirmation, the primary system manages the transmission and acknowledgment process. This intermediary role allows the VM to receive acknowledgment quickly while the background replication process continues independently, resolving the contradiction between reliability and latency.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If write-back approach is used without replication, then VM write latency is reduced, but data corruption and loss risk increases upon host failure

Engineering Contradiction:

ImproveVM write latencyVSAvoiddata loss risk

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent applies the copying principle by creating a replicated copy of the write-back cache on the secondary host system. When the primary host receives a write request, it transmits the data packet to the secondary system, which stores a copy. This copying mechanism ensures that if the primary host fails, the secondary system has a backup copy of the data, preventing data loss while maintaining the low-latency benefits of write-back caching.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent applies beforehand cushioning by establishing a backup replication mechanism before failures occur. The secondary host system is pre-configured and ready to receive data packets, acting as a cushion against potential primary host failures. This prior preparation ensures that data protection is in place before any failure event, allowing the system to maintain write-back performance while having safeguards ready to prevent data loss.

Inventive Principle:

Principle #11Beforehand cushioning (Prior cushioning)

Data Source

PatentUS9183103B2Lightweight remote replication of a local write-back cache

Publication Date: 2015.11.10 VMWARE INC

AI summary

Techniques for replicating a write-back cache are provided. In one embodiment, a first computer system can receive a write request from a virtual machine (VM) that includes data to be written to a shared storage device. The first computer system can further save the data in a local cache device and transmit, via a network interface controller (NIC), a data packet including the data to a second computer system. Upon determining that the data packet has been transmitted via the NIC, the first computer system can communicate an acknowledgment to the VM indicating that the write request has been successfully processed.