Industrial Manufacturing / Time Series & IIoTGHSense (Nanjing Guangheng)2026-07-18

GHSense: An Order-of-Magnitude Write Gain, and New Sites Without Custom Engineering

The bottleneck was the database, not the optics. Detection results were written to a hosted MySQL instance where tens of thousands of daily records were held together by sharding plus pre-aggregation. On MOI, write performance improved by an order of magnitude and the platform scaled past 70 devices without re-architecture.

GHSense (Nanjing Guangheng)

GHSense (Nanjing Guangheng Industrial Technology) builds high-precision opto-electronic sensing and inspection equipment for industrial manufacturers, combining laser sensing with machine vision and AI. Its core team previously worked on national programs including the Chang'e lunar missions and the Fengyun-4 meteorological satellites.

10x
Write performance gain
Milliseconds
Typical write latency
70+
Inspection devices
No sharding
Native date partitions

The challenge

The bottleneck was the database, not the optics. Structured detection results flowed from upstream software into a hosted MySQL instance, where tens of thousands of records a day accumulated until performance steadily degraded. As deployments grew to dozens of inspection devices, time-series write throughput hit a ceiling, and keeping queries usable required table partitioning plus pre-aggregation strategies-complexity that had to be re-engineered at every new site.

The solution

GHSense replaced MySQL with OmniFabric's converged engine. The partitioning-plus-aggregation workaround gave way to a native date-partitioned table design, so no sharding is needed as detection history accumulates, and serverless instances scale resources automatically with production load. More than 70 inspection devices now run on the platform.

The outcome

Write performance improved by an order of magnitude over MySQL, with most writes completing in milliseconds.

The majority of analytical queries over accumulated detection history complete within seconds, without the pre-aggregation the old stack required.

Native date-based partitioning replaced table partitioning plus aggregation strategies-new sites deploy without bespoke database engineering.

Solution Architecture

Data sources
  • Visual 2D Defect Detection (~60 Defect Types)
  • Production Line & Device Telemetry
  • Upstream Inspection Software Feeds
MatrixOne Intelligence
  • High-Throughput Time-Series Writes
  • Native Date-Partitioned Tables — No Sharding
  • Serverless Autoscaling with Production Load
  • Converged Engine: Writes and Analytical Queries Together
  • MySQL-Compatible Migration Path
Business applications
  • Real-Time Defect Records per Line
  • Historical Defect Trend Analysis
  • Per-Device and Per-Site Quality Reporting
  • Scale-Out to 70+ Devices Without Re-Architecture

Related solutions and products