MongoDB vs ReductStore: Choosing the Right Database for Robotics Applications

Robotics applications generate and process a wide variety of data, such as sensor readings, video streams, logs, and AI model outputs. Managing this data efficiently is crucial because it affects the performance, scalability, and reliability of the entire system.

In this article, we'll compare ReductStore and MongoDB, two databases designed to handle different aspects of data management. ReductStore is a time-series blob storage solution optimized for managing large amounts of data coming from continuous streams. MongoDB, on the other hand, is a popular NoSQL database known for its flexibility, scalability, and support for unstructured and semi-structured data.

By understanding the strengths and limitations of each, you can make an informed decision to meet your project's specific data needs.

TL;DR;

Feature	ReductStore	MongoDB
Architecture	Time-series string messages with metadata stored as key-value pairs	Collections of JSON-like documents
Data Querying	Query requests via time ranges with filters	MongoDB Query Language (MQL)
Communication Protocol	Uses HTTP/HTTPS	MongoDB Wire Protocol (binary, BSON-based)
Data Batching Capabilities	Supports data batching	Supports data batching, but it’s not optimized for time-series data
Replication	Granular data filtering support	Replicates the entire dataset; no native support for filtering during replication.
Retention	Retention policy based on data volume with ‘First-in, First-out’ (FIFO) deletions	Uses TTL (Time to Live) indexes for time-based data deletion
Performance	Optimized for unstructured data; up to 900% faster for writes and 2300% faster for reads.	General-purpose; slower with large blobs.

Understanding the General Architectural Differences

First, let's look at MongoDB and ReductStore's differences in terms of their general architectural principles.

ReductStore has a specialized approach to managing unstructured, “blob” data organized by time. Its architecture revolves around buckets, entries, and records—concepts comparable to databases, tables, and rows in SQL systems. This structure, combined with fixed-size blocks for chronological data storage, makes ReductStore highly efficient for real-time processing and retrieval of data coming from continuous streams. It's also important to be mentioned that ReductStore has no limits to the size or the type of the data you can store.

MongoDB, in contrast, uses a document-oriented model based on collections of JSON-like documents. This flexible structure supports complex hierarchies and adapts well to applications with dynamic or loosely structured data, such as e-commerce platforms or content management systems. While MongoDB has a 16MB document size limit, it can handle larger files using GridFS, a feature designed to store and retrieve large binary files by splitting them into smaller chunks. However, storing data with GridFS is slower compared to working with standard JSON documents, which can affect performance in applications requiring frequent or rapid access to large files.

Development Experience

In this section, we will compare the development experience of MongoDB and ReductStore, focusing on their query languages, communication protocols, and integration capabilities.

Query languages and SDK support

ReductStore supports querying data in several different ways – either by using its CLI, SDKs, or HTTP API. This database solution improves data retrieval by batching records within a time range, minimizing requests and delays. Apart from querying by a time range, there is a possibility to retrieve a record by its timestamp, but you have to keep in mind that such querying is less efficient due to the lack of batching. ReductStore provides client SDKs for Python, JavaScript, C++, and Rust.

MongoDB uses its own query language, MongoDB Query Language (MQL), which allows for flexible and powerful data querying. MQL supports complex filtering, aggregation, and indexing, making it a great choice for applications that need to work with semi-structured data. MongoDB also provides official SDKs for various languages, such as Java, Python, and C#, making it easy to integrate with various development environments.

Communication protocols

ReductStore uses HTTP/HTTPS protocols to handle data, focusing on reducing network load in environments with high latency. It processes large datasets by batching requests and delivering data in segments, making it efficient for both real-time and historical data access.

On the other hand, MongoDB uses its native MongoDB Wire Protocol, a lightweight, binary protocol optimized for fast communication between clients and servers. This protocol encodes data in BSON (Binary JSON), which supports efficient transmission and storage of JSON-like documents. These design choices make MongoDB a great choice for web-based applications, offering both flexibility and scalability.

Data Management Features

Data batching capabilities

Robots often generate high-frequency data from sensors and cameras, which requires efficient data transfer to avoid bottlenecks.

ReductStore enables efficient data batching by grouping multiple records into a single HTTP request, minimizing delays and reducing the number of queries needed. ReductStore optimizes data retrieval by batching multiple records into a single HTTP request, reducing network overhead and delays. The storage system ensures data records are batched optimally for the HTTP protocol, supporting use cases such as real-time monitoring and historical data analysis with minimal query overhead.

MongoDB, provides bulk write operations that enable inserting, updating, or deleting multiple documents in a single transaction. This is particularly useful for applications that need to modify large datasets without compromising performance. MongoDB's aggregation pipelines also support batch data processing, but this batching is more general and not a good option for time-series data.

Edge-to-Cloud Replication

In robotics, data is often processed locally (on edge devices) to reduce latency and dependence on cloud connectivity. However, important data still needs to be replicated to the cloud for centralized analysis, long-term storage, or to train machine learning models.

ReductStore offers granular control over data filtering during replication, making it easy to move specific subsets of data between edge devices and the cloud. This feature allows efficient syncing, ensuring only necessary data is transferred to reduce bandwidth usage. This makes ReductStore particularly well-suited for applications like robotics where data is processed in real-time at the edge, and only relevant or high-priority data needs to be replicated to the cloud for analysis or storage.

MongoDB's replication system mainly focuses on maintaining data consistency and fault tolerance across multiple replica sets. While it is very reliable and flexible for general-purpose applications, it is not specifically suited for edge-to-cloud workflows since it does not natively allow data filtering during the replication process.

Retention policies

Robotic systems generate large amounts of data, much of which quickly becomes irrelevant. Efficient retention policies are a must to maintain performance and avoid disk space issues on edge devices.

ReductStore's retention policies are based on volume thresholds, meaning that older entries are automatically removed as new ones are added. The platform follows a FIFO (first-in, first-out) rule, making it a perfect solution for applications where only recent data needs to be kept, such as sensor readings or telemetry logs. These retention policies can be customized for each data bucket, which ensures efficient data management.

MongoDB uses TTL (Time to Live) indexes to delete documents once they reach a certain age. This feature isn't automatic and requires setup, but it lets developers manage expiration policies for individual collections. TTL indexes are especially useful for applications like session storage, temporary logs, or data with defined lifespans.

Performance Analysis

MongoDB is often better suited for structured numerical data but requires a more complex setup with GridFS for handling blob storage. On the other hand, ReductStore simplifies the process by storing binary data directly into buckets, which makes it easier to manage time-series data.

In terms of performance, our analysis shows that ReductStore greatly outperforms MongoDB when handling large blobs of unstructured data. In write tests, ReductStore is up to 900% faster than MongoDB, while for read operations, it delivers up to 2300% faster results. In the table below, you can find the results of the tests conducted by measuring the speed of writing and reading of binary data of varying sizes on an NVMe drive.

Chunk Size	Operation	MongoDB, blob/s	ReductStore, blob/s	ReductStore, %
1 KB	Write	799	7994	+900%
	Read	2008	47979	+2300%
10 KB	Write	783	7431	+850%
	Read	1918	32888	+1600%
100 KB	Write	694	3612	+420%
	Read	1730	6250	+260%
1 MB	Write	246	663	+170%
	Read	776	540	-30%

Real-World Example

For a real-world robotics system, consider a scenario where numerous robots are operating on a manufacturing shop floor. These robots are continuously generating data, such as sensor readings, status updates, and environmental information. Kafka is used for ingesting this data in real-time and directing it into a local storage system for temporary retention and processing.

As the amount of data accumulates, it is very important to efficiently replicate this data to the cloud for further analysis and long-term storage. This is where ReductStore comes into play. The data replication strategy involves decreasing the volume of data by filtering out less important information automatically. By using ReductStore's edge-to-cloud replication and data filtering capabilities, only the most relevant time-series data is transmitted to the cloud, making sure that resources are used efficiently and bandwidth is optimized. This reduction strategy can be applied by using predefined rules or dynamic triggers, such as certain events or changes in sensor readings, helping improve performance and save on cloud costs.

Conclusion

ReductStore and MongoDB each have their own strengths, making them a good fit for different needs and use cases. MongoDB is great for flexible, general-purpose storage, especially with structured data, but when it comes to handling large-scale, time-sensitive data—like the kind found in robotics—ReductStore really shines. With optimized performance, fast read/write speeds, and features like edge-to-cloud replication, ReductStore is better suited for managing large, unstructured data sets. So, while MongoDB is a solid choice for many use cases, ReductStore stands out for robotics applications that need efficient, real-time data management.

---

I hope this tutorial has been helpful. If you have any questions or feedback, don't hesitate to use the ReductStore Community forum.