Copyright (C) 2014-2015 Jaguar Land Rover
This document is licensed under Creative Commons Attribution-ShareAlike 4.0 International.
This document gives a brief introduction to the codebase of the RVI project and explains the reasoning behind some of the technical choices.
For a high level description, with an exhaustive master usecase walkthrough, please see the High Level Design document here
Git branch management is JLR OSTCs standard git document Git strategy Git strategy
For build instructions, please check the build instructions: Markdown | PDF
For configuration and launch instructions, please check the configuration documentation: Markdown | PDF
Technical RVI disussions are held at the GENIVI project mailing list: GENIVI
GENIVI RIV Expert Group (RVI-EG) discussions are held at the members only list: GENIVI
The Remote Vehicle Interaction project will specify, design, plan and build a reference implementation of the infrastructure that drives next generation's connected vehicle services.
-
Specify
Requirement specifications, test suites, integration tests. -
Design
High Level Description. Detailed Description. Use Cases. -
Plan
Roadmap. Milestones. Deliverables. Budgeting. Resource planning. -
Build
Implement. Document. Test. Demonstrate. Deploy for download. -
Reference Implementation
Provides a baseline and starting point for organizations' production-grade connected vehicle projects.
RVI provides P2P based provisioning, authentication, authorization, discovery and invocation between services running inside and outside a vehicle.
-
P2P
Internet connection not required for two peers to exchange services. -
Provisioning
Add, delete, and modify services and network nodes. -
Authentication & Authorization
Proves that a service is who it claims to be, and has the right to invoke another service. -
Discovery
RVI is considered a safe bet that can be used without unforeseen consequences. -
Invocation
RVI shall be able to function over transient and unreliable data channels, but also over a reliable in-vehicle LAN.
The following chapters describe the technology choices made for the reference implementation. An overall goal was to avoid technology lock-in while easing adoption by allowing organiztions to replace individual components with in-house developted variants using their own technology.
The only fixed parts of the entire RVI project are the JSON-RPC protocol specifications used between the components themselves and their connected services.
GENIVI provides an RVI implementation as a starting point and a reference. The adopting organization is free to use, rewrite, or replace each component as they see fit. A typical example would be the Service Discovery component, which must often be extended to interact with organization-specific provisioning databases.
In a similar manner, RVI places no requirements or expectations on the communication protocol between two nodes, such as between a vehicle and a backend server, in an RVI network. An organization is free to integrate with any existing protocols, or develop new ones, as necessary.
The reference implemantation provides an example of how to handle a classic client-server model. The RVI design, however, can easily handle such cases such as wakeup-SMS (used to get a vehicle to call in to a server), packet-based data, peer-to-peer networks without any dedicated servers, etc. The adopting organization can implement their own Data Link and Protocol components to handle communication link management and data encoding / decoding over any media (IP or non-IP).
Erlang was chosen to implementation the core components of the RVI system. Each component (see the HLD for details) run as an erlang application inside a single erlang node.
Several reasons exist for this somewhat unorthodox choice of implementation language:
-
Robustness
Erlang has the ability to gracefully handle component crashes / restarts without availability degradation. This makes a deployment resiliant against the occasional bug and malfunction. In a similar manner, redundant sites can be setup to handle catastrophic failures and geographically distributed deployments. -
Tool availability
There are a multitude of open source erlang components available to handle SMS, GPIO, CAN buses, GPS, PPP links, and almost any protocol out there. Since erlang was designed to handle the mobile communication requirements that is at the center of the connected vehicle, integrating with existing systems and protocols is often a straight-forward process. -
Scalability
The concurrent nature of an erlang system sets the stage for horizontal scalaility by simply adding hosts to a deployment, allowing an organziation to expand from a pilot fleet to a full fledged international deployment. -
Carrier grade availability
The robustness and scalaiblity, in conjunction with the built-in erlang feature of runtime code upgrades, is a part of erlang's five-nines uptime design that is rapidly becoming a core requirement of the automotive industry. -
Proven embedded system solution
Erlang has been adapted to operate well in embeded environments with unreliable power, limited resources, and the need to integrate with a wide variety of hardware.
Python is used to implement all demonstrations, beginning
with the HVAC demo available in the hvac_demo subdirectory.
By using Python for the demos, which is better known than erlang, examples are given on how to write applications and services interfacing with the RVI system.
Performance is not a goal of the RVI reference implementation. Instead, code readability and component interchangeability takes priority in order to ease design understanding and adoption.
One example is the use of JSON-RPC (over HTTP) to handle internal communication between components in a single Erlang node.
Using a traditional erlang solution such as genserver, the overhead for internal transactions could be cut to a few percent in comparison with the current JSON-RPC implementation. That route, however, would force all components of the RVI system to be implemented in Erlang, thus severly limiting an organizations abilities to replace individual components with their own versions.
All code in the RVI reference implementaion and its demonstrations are written with a minimum of complexity and "magic". Readability is paramount, even if it severly impact performance and memory usage.
All components in the RVI are kept small and distcinct, with a well-defined JSON-RPC external interface and a simple call flows.
Only three external modules (lager, bert and exo) are used by the code, with two more (setup and edown) used for release and documentation management.
The reason for minimizing external module usage is to make the code comprehensible and minimize the time a developer has to travesrse through obscure libraries trying to understand what a specific call flow actually does.
The entire reference implementation (as of the first alpa release) is 2800 lines of code, broken down into six standalone modules and one library of shared primitive functions.
JSON-RPC is used for all communication between components in an RVI system, and also to communicate with services connected to it. The ubiquity of JSON-RPC, and its close relationship with Java/Javascript, provides maximum of freedom of technology choices when new components, services, and applications are developed.