ReductStore

Want ROS 2 logs that don’t eat your disk or break your pipelines? Here’s the formula we use for fast, simple, scalable storage. 📦 Use MCAP to log structured ROS 2 data ⏱️ Split recordings into 1–5 minute files (by time or size) 🗂️ Consider grouping related topics into separate files ♻️ Set a FIFO quota to auto-delete old files when space runs low 🧠 Apply chunk-level compression to save disk (uses some CPU/mem); but avoid file compression Keeps logging simple and scalable; from the robots to the cloud. 🔗 https://lnkd.in/e8mqdAme #ros #robotics #database

Getting a streaming stack running should be simple. We use three tools: Zookeeper, Kafka, and ReductStore. Each does one thing well—coordination, event streaming, and time-series blob storage. Put them together and you get a solid base for projects like anomaly detection, computer vision, and robotics. No extra pieces. No feature bloat. Kafka moves data fast, but keeping raw data long-term is a hassle. ReductStore fills that gap—just store whatever you need, keep it timestamped, and pull it out later. We don’t waste time wiring up complicated systems. We just connect the basics and get to work. If you want to see a simple setup, let me know. The goal is clear: less time on plumbing, more time on solving the real problems: https://lnkd.in/eaZCK36b #DockerCompose #KafkaTutorial #DevOps"

Everyone obsesses over faster computer vision models. But almost nobody talks about the real problem: Not your model. Not your GPU. 📦 It’s your storage crying in the corner. Computer vision systems don’t just need fast inference. They need somewhere to put all that data. 🧠 Real-time processing 📉 Smart compression 📍 Edge-first thinking These aren’t “nice to have”, they’re survival tactics. Because let’s be honest: If your storage can’t keep up, You're just generating data you can’t use. #highfps #computervision #datamanagement

Victor Kataev, thank you for the awesome contribution!

Power, precision, and a seriously capable platform for real-world robotics! The Unitree A2 is a big step forward. More torque, better balance, and full-body control at 10 kHz. That opens the door to tougher terrain, heavier payloads, and more dynamic behavior. The next challenge is everything behind the scenes: capturing high-frequency sensor data, keeping it structured, and making sure it's accessible for training, analysis, and debugging. That’s where we come in. Great release from the team at Unitree Robotics.

Most robotics teams don’t notice their storage problem until it’s too late. At first, rosbag works fine. You log some sensor data, maybe a few hours of video. No trouble. But give it a year: Now you’ve got terabytes of data, packed away in bag files you never want to touch again. And when you need to troubleshoot, or run ML experiments, or just find that one five-minute window from last October—it’s painful. Rosbag was never designed for long-term, searchable storage. Backing up to cloud buckets is just another headache. That’s where time-series object storage makes a difference. Instead of dumping everything in giant files, you get: - Cheap, scalable storage built for streaming data - Fast search and retrieval based on time - Easy integration with analysis tools and cloud workflows We built ReductStore for keeping logs, LiDAR, camera streams.. To grab exactly the data you need, when you need it—no downloading fifty gigabytes or parsing weird file formats. Cost drops, access get easier, and the storage just scales. If you’re still living with rosbag archives, it might be time to try object storage designed for time-series data. #ROS #Robotics #DataManagement

Real-world deployments like this show that autonomy lives or dies by the quality of its data infrastructure. In airport environments, robots don’t just need sensors. They need reliable edge storage, smart filtering, and fast replication to the cloud when bandwidth allows. This is the kind of environment where ReductStore thrives. Credit to Aéroport de Bordeaux for pushing the frontier.

⚙️ TSDBs like InfluxDB are fast (but expensive). ReductStore gives you both speed and savings. Use low-cost Blob storage ($20/TB/month) to reduce costs by tenfold compared to traditional TSDBs, such as TimescaleDB and InfluxDB. This setup maintains comparable or superior performance for record sizes exceeding 1kB. Explore ReductStore to optimize your storage costs and performance. 🔗 https://www.reduct.store/ #DataLakehouse #TimeSeriesDatabase #Manufacturing

The ocean floor is becoming one of the most data-rich environments on Earth, but only if you can handle the flow. As Kraken Robotics highlights, underwater LiDAR isn't just about better visuals. It's about creating high-fidelity digital twins. These demand real-time processing, long-term storage and structured data workflows. From subsea robotics to 3D mapping, precision only scales when your data pipeline scales too.

🕒 MQTT wasn’t built to store months of data. It’s a great protocol for lightweight, real-time messaging between IoT devices; but it’s not meant for storage. Here’s the problem: Most MQTT brokers have limited in-memory buffers or short-lived persistence settings. You might be capturing critical telemetry; but unless you’ve added a real storage layer, that data is at risk. A power outage, a device reboot, or a misconfigured retention setting... and poof 💨 it’s gone. So what’s the solution? For long-term storage you might need a time-series object store: ✅ Store full message payloads ✅ Organize by topic and timestamp ✅ Query, replicate, and retain for months or years We wrote a simple Python tutorial to show you how: 🔗 https://lnkd.in/eDeRhB4t #MQTT #TimeSeriesDatabase #IoTData