In a recent conversation with Dmitry Alexandrov, DBA Team Leader at Exness, we explored how MinIO enables Exness to handle massive data volumes and deliver lightning-fast trading experiences. Exness is a global leader in online trading, known for its transparency, technological innovation, and customer-centric approach. Its trading platform generates and processes petabytes of data, requiring a cutting-edge infrastructure to ensure seamless operations.

Dmitry shared insights on Exness’ data challenges, its decision to move from clouds to an on-prem infrastructure with MinIO, and how MinIO has become central to their AI/ML workloads, disaster recovery, and more.

Scale and Savings: The Only Solution

When asked about the storage challenges Exness faced, Dmitry described a perfect storm of exploding data volumes and the need for high throughput:

“We started to create our data lake project about a year ago, starting from 200 terabytes. Now we have almost half a petabyte in our data lake in MinIO.”

Exness’ storage journey began with clouds, but as its infrastructure scaled, so did its costs. Dmitry explained the rationale behind moving to MinIO on-prem:

“After calculations, we decided it would be a more cost-effective solution to host object storage and all compute connected to it in our on-prem infrastructure.”

MinIO’s S3 API Compatibility

Exness’ extensive use of clouds meant their workflows were deeply tied to the S3 API. Migration could have been a nightmare—if not for MinIO.

“Our primary requirements when we started this project were strong S3 compatibility because we are leveraging many existing tools and workflows that depend on the S3 API since we used AWS before.

There was only one object storage solution that offered a very high compatibility with the S3 API with performance and scale: “MinIO really stood out. It not only met our requirements but also offered an open-source model that allowed us to customize and scale on demand efficiently.”

AI and ML Workloads: MinIO at the Core

Exness’ focus on AI and machine learning continues to grow, Dmitry shared how MinIO powers their AI/ML workflows:

“MinIO actually plays a central role not only in our data lake. In several of our applications, we run a lot of Spark jobs that are orchestrated by Airflow. These jobs often require fast access to data storage in object storage, and MinIO enables that with high speed and reliability.”

Exness is also building a machine learning model registry using DVC and MLflow, with MinIO at its core:

“We are planning to expand with a machine learning model registry built on DVC and MLflow, which will also use MinIO even though it is in development. This registry will serve as a centralized hub for all ML models, making it easy for our teams to manage and deploy them seamlessly.”

Handling Extreme Workloads with Ease

One of the standout moments from our conversation with Dmitry was his description of MinIO’s ability to handle extreme traffic spikes:

“Just last week, we hit a traffic peak from our cluster to 200 gigabits per second, with one user alone generating 170 gigabits per second. Despite those volumes, all this traffic was handled effortlessly.”

Disaster Recovery and Replication

Data reliability is critical in the financial sector, and Exness takes disaster recovery seriously. MinIO’s replication features play a key role:

“We replicate all the data to another data center. In case of disaster recovery, we can switch to the other data center and serve the workload from there.”

Flexibility Beyond the Data Lake

In addition to MinIO powering Exness’ data lake, Dmitry highlighted its versatility across other use cases:

“We also use MinIO as our GitLab registry. We use it for document storage. We use it as backup storage as well. This flexibility has made it invaluable across a wide range of needs.”

Looking Ahead: Scaling AI and Data

Exness’s plans for the future are ambitious. Dmitry shared what’s next for their MinIO-powered infrastructure:

“We are doubling our data lake capacity and investing heavily in AI/ML. MinIO will remain at the core, especially as we roll out our model registry and train more complex models. We’re also enhancing security with advanced encryption and access controls to meet the highest standards.”

Lessons from Exness’ Journey

Exness’ story offers valuable insights for organizations managing large-scale data infrastructures:

Scale On-Prem: At scale, colocating compute and storage with MinIO on-prem offers unmatched performance and cost savings.
Prioritize Compatibility: MinIO’s close compatibility with the S3 API ensures seamless integration in the data stack, reducing operational friction.
Invest in Versatility: MinIO’s ability to support data lakes, AI/ML workloads, advanced analytics and general-purpose storage of both structured and unstructured data makes it an essential tool for modern data infrastructure.

Closing Thoughts

Dmitry closed the interview by reflecting on MinIO’s impact on Exness:

“It’s not only about trading data. It’s data for analytics and for machine learning purposes. MinIO delivers the performance and reliability we need.”

Exness’ success demonstrates how MinIO empowers organizations to scale their infrastructure without compromise, paving the way for innovation in AI, analytics, and beyond.

Tune in to the podcast to listen to our conversation here.