Overview of JDNoSQL Platform and Its Real-Time Advertising Use Cases

JDNoSQL is a distributed, column‑oriented key‑value storage service built on HDFS that provides millisecond‑level read/write, dynamic columns, horizontal scalability, high concurrency, low latency, high availability, and strong consistency, supporting both structured and unstructured data.

The platform is positioned as a NoSQL solution for large‑scale, real‑time random access workloads, addressing the limitations of traditional relational databases in handling massive, distributed datasets.

Key characteristics of JDNoSQL include:

Massive scale: tables can contain billions of rows and millions of columns.

Column‑oriented storage with independent column retrieval.

Sparse storage: NULL columns consume no space.

Schema‑less design: rows have a sortable primary key and can have dynamic columns.

Multi‑version data with timestamp‑based version numbers.

All values stored as strings.

Typical application scenarios at JD.com cover time‑series services, messaging, real‑time analytics (CUBE), monitoring, feed streams, AI storage, spatio‑temporal data, and financial analytics.

One detailed use case is a real‑time advertising computation system, which processes four main data types: ad impressions, ad clicks, ad actions, and third‑party monitoring data. Sample data formats are shown below:

2015-01-13 19:11:55{00D81D1D-00A291-0E2300-87DBCE0DA90} {"adia":"31769","asid":"2","aspid":"0","ptime":"14","ag":"4,5.20,26.1908","ecode":"15","type":"2","dp1":"1","adpid":"0","dsp":"0","source":"s"}61.237.239.3 天津 天津市

2015-01-13 00:11:06{D33333C3-000C84-2345FB-DB768EC56} {"wid":"13","aid":"103297","vid":"1446779","adid":"29260","asid":"1","aspid":"1","mid":"16507","mg":"155","area":"13","dsp":"3"} 175.8.146.246 湖南省 长沙市

The system must process these streams in real time, handling challenges such as unordered arrival, infinite stream size, and the need for immediate processing or archiving.

The architecture is divided into six layers: Log Reception, Producer, Message Queue (Kafka), Consumer, Business Logic (Flink), and Storage (JDNoSQL). This design enables high‑throughput, low‑latency processing with TPS up to 5 million requests and end‑to‑end latency under 30 seconds.

Table designs for real‑time ad statistics are provided. The impression table ( realtime_adpv) uses composite row keys like addid_省名称_20150113 with a single column family pv and column cnt. The click table ( realtime_adclick) has a simpler row key adid_20150113 with column family clk and column cnt. Example rows are shown with hexadecimal byte values representing long integers.

These tables support various real‑time queries, such as current ad delivery per province, city, or user, historical trends over a month, and click volume trends. Front‑end visualisation can be built with libraries like ECharts, Highcharts, or D3.js.

In conclusion, NoSQL databases are projected to grow rapidly, with JDNoSQL offering a comprehensive ecosystem that includes APIs, PhoenixSQL, OpenTSDB, Solr/Elasticsearch, GeoMesa, HGraph, and Spark on HBase, meeting diverse business needs.