A Database Optimized for Telcos

Stratum’s requirements as a distributed data store for telecoms are clear: high availability must be guaranteed and is a non-negotiable requirement, while low client response latencies should be prioritized. Within the PACELC framework, this design represents a deliberate choice for Stratum to prioritize Partition tolerance and Availability (PA) during failures and Latency over strict Consistency (EL) during normal operation.

Attempting to enforce strict real-time consistency across all replicas introduces additional coordination overhead on every transaction, increasing response latency and cross-replica communication. Sustaining such a model demands greater compute, memory, and network capacity to absorb these costs.

However, even with substantial infrastructure investment, the latency cost cannot be eliminated; it is intrinsic to the consistency model itself. In practice, for most types of data, the small possibility of a stale database read is less costly than the performance penalties and architectural rigidity imposed by strict consistency.

Grounded in this reasoning, Stratum adopts a model of practically bounded eventual consistency: write requests are accepted locally to preserve responsiveness and availability, while background replication mechanisms ensure that data automatically and reliably converges across datacenters. The system is engineered so that convergence is guaranteed and operationally transparent even following failures, eliminating the need for manual intervention.
Over the past decade, many globally distributed platforms have intentionally adopted similar designs to achieve scale, high availability, and low latency. Platforms such as Netflix and Twitter (X) operate large-scale distributed systems built on eventually consistent storage models to support massive write throughput and continuous operation across geographic regions.

Practical studies and production usages have shown that when designed correctly, eventually consistent systems behave in a way that is as predictable and reliable from the client’s perspective as a strictly consistent system. At the same time, they deliver the key benefits required by large-scale services: uninterrupted availability, low-latency local writes, and efficient horizontal scaling across geographically distributed datacenters.[8] That is the lens through which Stratum’s architecture was founded and has continuously evolved in close cooperation with world leading telecom operators, scaling to store hundreds of millions of records while servicing millions of transactions per second.

Instead of adapting general-purpose databases, Enea’s Stratum implements a purpose-built document database engine designed from the ground up as a telecom network data layer, with an architecture design that delivers three outcomes:

• High Availability through Write-anywhere: clients can write to any site, avoiding wide-area-network latency in the write path and eliminating complex leader/ownership failover behavior, with deterministic conflict resolution ensuring safe convergence.
• Horizontal scalability through push-based data replication: replication has a fixed cost and replication lag is effectively equal to the inter-site network latency.
• Operational sustainability through continuous data anti-entropy: a background process reconciles missed updates, network partitions, process restarts, and new-site bootstrap – ensuring data convergence without operator intervention.

This article explains why the Stratum design goals align with the real trade-offs for operators planning large-scale telecom deployments.