In Progress Engineer Documentation

Eclipse Xpanse is an open-source framework designed to build, deploy, and move managed services across multiple public and private clouds. It abstracts the complexities of cloud services by providing a standardized approach to the service catalog, life-cycle management, deployment, monitoring, and metering. It is particularly useful for small and medium cloud providers to offer native managed services on any OpenStack-based cloud.

Project Eclipse Xpanse is a framework for building and managing managed services across multiple cloud environments, based on the Eclipse Open Services Cloud (OSC) initiative. It abstracts the complexities of cloud services by providing standardized mechanisms for service catalogs, life-cycle management, deployment, monitoring, and metering. It is compatible with large hyperscalers as well as small to medium providers, specifically supporting any OpenStack-based cloud, SCS clouds, HUAWEI cloud, and Flexible Engine.

Eclipse Xpanse and Sovereign Cloud Stacks (SCS) provide a combined open-source framework for cloud infrastructure and services designed to ensure digital sovereignty and prevent vendor lock-in. Eclipse Xpanse serves as the open-source framework for cloud services, enabling services to move across any cloud provider. Sovereign Cloud Stacks (SCS) provides an open-source cloud stack built from upstream projects of the Linux Foundation (LF), Cloud Native Computing Foundation (CNCF), and Open Infrastructure Foundation (OIF), allowing cloud providers and private operators to deploy certified standards for core workloads.

LSM-trees are data structures optimized for write-heavy workloads, contrasting with B-Trees which are optimized for read-heavy workloads. The architecture is defined by two main characteristics: it is 'Log-structured' because transactions are appended to a memory table (mem-table), and it is a 'Merge-tree' because data is subsequently merged into Sorted String Tables (SSTables) on disk.

The JVM garbage collector performs compaction to resolve memory fragmentation. During this process, the collector moves objects closer together in the heap and updates all existing references to those objects. This results in larger contiguous blocks of free memory, which improves the speed of memory allocation and reduces fragmentation. This process is transparent to the application; the JVM manages the memory address changes internally, which is why the JVM does not expose raw memory addresses to the application.

A Stream is the actual input or output source of bytes (the data flow). A Buffer is a region of continuous memory used to store binary data temporarily from a stream before it is processed or moved to its next destination. For example, data coming from a disk or network streams into a buffer, which can then be read and converted into a specific encoding.

A cache line is the fixed size of a cache entry, typically 64 bytes. This value is determined by the CPU manufacturer and cannot be configured by the user. The entire cache area is divided into these fixed-size blocks. All data reading and writing occurs via the CPU cache (L1, L2, L3) based on the principle of spatial locality—the assumption that data requested now will likely be needed again soon.

Chat templates are a feature of the AI model, defined during its training and included as part of the model extract. They define the specific structure the model expects for input. When using an inference engine, the engine is responsible for converting standard requests (such as those using OpenAI-compatible HTTP APIs and data models) into the specific format required by the model's chat template before processing.

The odometer model explains how all number systems function: each position can hold a value from 0 up to (base - 1). When a position is incremented past its maximum value, it rolls over to 0, and the next position to the left increases by 1. For example, in binary (base-2), the allowed values are 0 and 1, resulting in a sequence like 0, 1, 10, 11, 100, etc.

Modular arithmetic is a system of arithmetic for integers where numbers 'wrap around' upon reaching a certain value (the modulus). It is often referred to as 'clock arithmetic'. In a system of modulo N, the only possible values are 0 to N-1. While the mathematical operation is performed by dividing and taking the remainder, the conceptual mental model is a counter moving in a circle; the result is what remains on the circle after the final round.

In Java, functions cannot exist independently of classes. Anonymous classes provide a way to create function objects. An anonymous class acts as a function object specifically when it implements exactly one method; if it implements multiple methods, it is considered a regular object. While functional, anonymous classes are more verbose than lambdas.