July 2026 Summaries

OpenTelemetry architectures typically involve applications emitting generic telemetry, which collectors batch and export to a managed backend for processing and storage. This approach works well for most scales, allowing backend systems to handle data parsing, encoding, and storage without requiring changes from the applications. However, at petabyte-scale, the ingestion process can become costly, leading to inefficiencies in data handling. Axiom's solution, Bifrost, addresses this by shifting data preparation closer to the source, applying Axiom-specific encoding and storage logic before the data reaches the central system. This approach reduces the computational burden on central systems and lowers data-loading costs by producing compact columnar batches that are ready for storage and querying. While still in private preview, Bifrost is designed to optimize the economics of large-scale data telemetry by maintaining data integrity and enabling efficient query processes, thus providing a more cost-effective solution for handling massive volumes of machine data.

Bifrost is a new data-loading path for Axiom's machine data platform, designed to significantly reduce the cost of loading OpenTelemetry data at petabyte scale by leveraging Axiom's expertise in encoding, compression, and batching closer to the data source. Unlike the conventional centralized pipeline that processes every byte after data arrival, Bifrost prepares query-ready columnar batches near the data source, allowing for immediate streaming into object storage and subsequent querying without additional processing. This approach eliminates the need for ingest compute, thus reducing costs associated with data reshaping and bandwidth, and it supports keeping more OpenTelemetry data by minimizing the need for data sampling. Bifrost remains aligned with Axiom's managed platform, ensuring that data retains the same storage, retention, and query capabilities as the existing pipeline, with loading costs starting significantly lower than the standard path. Currently in private preview, Bifrost is aimed at users managing large-scale OpenTelemetry workloads, promising considerable savings and efficiency in handling massive data volumes.

OpenTelemetry (OTel) is a comprehensive set of APIs, libraries, and services designed for capturing distributed traces and metrics, which is particularly advantageous for microservices and serverless architectures due to its standard-based telemetry data collection. Supported by the Cloud Native Computing Foundation, OTel is rapidly becoming the preferred standard among platform engineers and DevOps professionals, offering a unified API for instrumentation and interoperability across various infrastructure components. Axiom enhances the power of OTel by capturing data at full fidelity without conversion or sampling, allowing for seamless query transformation through its APL language, which is designed to be more intuitive than traditional SQL. This empowers non-technical team members, such as product managers, to perform their own queries without requiring engineering support. Axiom's pricing model starts at $25 per month, ensuring cost-effective access to OTel capabilities without unexpected expenses.

The text discusses the Gorilla algorithm, initially developed by Facebook for time series caching and storage, highlighting its simplicity and effectiveness in handling monotonically increasing timestamps and float-only, append-only data. While Gorilla has become a standard, it presents challenges in multi-tenant deployments, particularly with out-of-order data and integer support. The author describes an alternative approach that introduces two buffers (compressed and uncompressed) to manage out-of-order data and employs a bitmap for time storage, which improves compression efficiency over Gorilla's delta-of-delta method. This new method, compatible with both integers and floats, trades some CPU efficiency for better memory and storage management, achieving a significant reduction in storage size while being slightly less memory efficient. The approach is tailored to handle both floats and integers, addressing shortcomings in Gorilla's handling of data types and order, ultimately providing a net win in storage efficiency despite increased CPU usage.