Mogulinker Technology (Mogulinker) is leading the charge in industrial energy-saving and digital transformation with its innovative Lingzhi AI model, designed to optimize high-energy-consuming equipment and auxiliary energy workshops. This collaboration with TDengine brings their strengths together to create an AI-powered smart control platform, revolutionizing energy management in manufacturing. This article explores why Mogulinker chose TDengine, the challenges faced during implementation, and the shared vision driving their partnership forward.
Traditional manual control methods in industrial settings often result in inefficiencies like continuous device operation, poor parameter settings, and redundant energy supply, leading to significant energy waste. While lean management can help, the real solution lies in adopting data-driven digital tools to maximize data value, cut costs, boost productivity, and enhance quality.
Mogulinker is an industrial AI company specializing in high-energy-consuming equipment and auxiliary energy workshops. With its proprietary Lingzhi AI model, Mogulinker leverages AI to address redundant energy supply in factories by analyzing and optimizing energy supply and demand data. This enables factories to achieve on-demand energy usage, ensuring safe energy supply, automated operation, and sustainable energy savings with reduced carbon emissions.
Mogulinker’s cloud smart control platform provides energy management and optimization across four levels: device, workshop, plant, and group. It excels in high-energy-use areas like air compression and cooling stations while enabling predictive maintenance through data insights. The platform’s modular design allows flexible deployment for specific needs or comprehensive coverage of energy types such as water, electricity, gas, and cooling. Currently, it supports over 1,600 industrial enterprises, handling around 100GB of IoT data daily with a multi-tenant model.
As Mogulinker drives digital transformation for industrial enterprises, it encounters both common industry challenges and unique business demands. According to its R&D lead, the platform’s data storage must be decoupled from specific business scenarios and support dynamic field definitions. In auxiliary energy settings, the diverse range of equipment and brands generates over 100GB of IoT data daily, posing significant challenges for data management and processing. The core requirements are threefold:
- Parameter codes for similar devices can vary, with some parameters being specific to certain models. The system must dynamically store and analyze these unique parameters while allowing for the seamless integration of new device types by adapting the data structure as needed. This ensures all device data can be fully captured and integrated into the existing system.
- With daily data volumes exceeding 100GB, the system must deliver response times within 200 milliseconds, requiring exceptional real-time query performance and high availability.
- Clients need flexibility for both public and private cloud deployments. The database must support small-scale private setups as well as large-scale cloud clusters, ensuring consistent development and operational efficiency across different environments.
After testing databases like OpenTSDB, HBase, InfluxDB, and others, Mogulinker selected TDengine for its superior performance, stability, and data compression capabilities.
Virtual Table Structure Design
The AI Cloud Smart Control platform needs to process data from a wide variety of devices, some with thousands of code fields and undefined data structures. Under these circumstances, Mogulinker could not utilize TDengine’s Supertable model due to its column limitations and inability to handle uncertain field structures.
To address this, Mogulinker implemented TDengine’s recommendation to use a regular table model, mapping each device’s code to a sub-table ID and column name. This approach enables schema-less storage at the device level and bypasses column limitations.
High-Concurrency Data Query Performance
“Each tenant generates and processes large volumes of business metrics, using both streaming and batch modes to write data to the time-series database. While basic calculations are handled within the database, more complex computations are performed in the business system’s memory. This demands high performance, efficiency, and stability from the database to handle complex data ingestion and querying requirements.”
Mogulinker tested the performance of TDengine (v.3.2.3.0) against InfluxDB OSS v.1.8 and v.2.7 on a single machine with an 8-core CPU and 32GB of memory to evaluate the database’s capabilities.
| Query Granularity | Time Range | TDengine | InfluxDB 2.7.6 | InfluxDB 1.8 | |||
|---|---|---|---|---|---|---|---|
| Query Details | 7 days | 1790 | 1728 | 106 | 90 | 137 | 145 |
| 3m Downsampling first | 7 days | 24 | 21 | 113 | 114 | 93 | 90 |
| 1h Downsampling first | 7 days | 5 | 4 | 12 | 14 | 25 | 30 |
| 1d Downsampling first | 7 days | 3 | 3 | 6 | 7 | 20 | 24 |
| 3m Downsampling first | 6 months | 589 | 608 | 3680 | 3183 | 1784 | 1903 |
| 1h Downsampling first | 6 months | 37 | 36 | 233 | 217 | 152 | 208 |
| 1d Downsampling first | 6 months | 13 | 14 | 84 | 84 | 100 | 97 |
| 3m Downsampling diff | 7 days | 23 | 19 | 124 | 115 | 92 | 89 |
| 1h Downsampling diff | 7 days | 3 | 3 | 12 | 14 | 29 | 28 |
| 1d Downsampling diff | 7 days | 3 | 3 | 10 | 6 | 23 | 27 |
| 3m Downsampling diff | 6 months | 405 | 407 | 3426 | 3229 | 1811 | 1869 |
| 1h Downsampling diff | 6 months | 32 | 32 | 255 | 219 | 203 | 185 |
| 1d Downsampling diff | 6 months | 11 | 11 | 85 | 64 | 100 | 98 |
Overall, TDengine outperformed InfluxDB OSS v.1.8 and v.2.7 in most aggregation queries, delivering 3-10 times faster performance and meeting our requirements. Compared to HBase and InfluxDB, it reduced response times for complex queries from seconds to milliseconds, significantly improving report performance and enhancing the user experience.
Cloud and Private Deployment Consistency
TDengine’s C++ design supports both single-machine performance and cluster deployment, even in resource-constrained environments, enabling consistent architecture for both public and private cloud deployments.
Efficient Handling of Large Volumes of Metrics and Queries
In Mogulinker’s typical use cases, IoT data undergoes multi-dimensional processing to generate a wide range of metrics tailored to different business needs. These include electricity and cost calculations, carbon and coal conversions, equipment runtime statistics, utilization analysis, single-device energy efficiency evaluations, air-compression energy ratios, central air-conditioning COP, energy consumption per product unit, productivity per monetary value, pressure and flow forecasts, and energy-saving rate calculations.
Some metrics are stored directly through stream-batch processing, while others require additional processing during queries. To handle these scenarios, Mogulinker developed custom built-in functions to address diverse business needs. These demands pushed the time-series database’s write and query efficiency to its limits, and after thorough validation, TDengine delivered excellent performance in both data ingestion and querying, fully meeting Mogulinker’s requirements.
