Sammy Kerata Oina
55afe4c038
* Optimize QEMU launch and add V-sock support Refactored QEMU argument construction and launching logic by removing the dependency on 'agent.Computation'. This simplification makes the VM creation process more streamlined. Additionally, introduced V-sock capabilities in the QEMU configuration to facilitate improved guest-host communication. Updated the README to include kernel module setup instructions for the new V-sock feature. The V-sock implementation enables VMs to use a consistent communication channel that is not affected by network configuration changes, enhancing reliability and potential interoperability with host services. It's important to ensure that the necessary kernel modules are loaded as part of the setup process, as documented. Signed-off-by: SammyOina <sammyoina@gmail.com> * Add vsock-based communication to manager Introduced virtual socket (vsock) communication abilities in the manager package by implementing a new socket service. This includes establishing a vsock listener and stub methods for sending computation results and cleaning up resources. The addition provides the groundwork for interprocess communication between guest and host in virtualized environments. - Integrated the `mdlayher/vsock` library for handling virtual socket operations. - Created a new `sockService` struct to encapsulate vsock listener handling. - Implemented `NewVsock` constructor to initialize the listener with domain value `3`. - Added placeholder methods for future computation sending and service closing logic. This enhancement targets scenarios where efficient VM-to-host communication is required. Signed-off-by: SammyOina <sammyoina@gmail.com> * remove env Signed-off-by: SammyOina <sammyoina@gmail.com> * Refactor agent config and use vsock Introduce `AgentConfig` struct to group agent-related configurations, and update `Computation` struct to include the new `AgentConfig` field. Replace command-line computation extraction with vsock-based config retrieval for robustness and decoupling. The agent configuration is now read from a vsock connection during runtime, allowing for more dynamic and flexible deployments. Adjusted the main agent application logic to support these configuration changes, and corresponding changes have been made in the manager to facilitate vsock communication. This approach aligns with modern practices for microservices by streamlining configuration management and reducing reliance on static command-line parameters. Moreover, it enhances the scalability of the agent service by allowing configuration to be managed externally. Signed-off-by: SammyOina <sammyoina@gmail.com> * Refactor agent config and remove deprecated code Consolidated agent configuration management into a single `AgentConfig` message and pruned deprecated Protobuf `ComputationReq`, `DatasetReq`, and `AlgorithmReq` messages. Adapted corresponding manager service logic to the new configuration structure. These modifications align with updated manager API schema, facilitate clearer configuration handling, and improve maintainability. Signed-off-by: SammyOina <sammyoina@gmail.com> * send configuration Signed-off-by: SammyOina <sammyoina@gmail.com> * Switch agent to listen mode for manager connections Previously, the agent established a connection to the manager using a direct dial. This change shifts the setup to where the agent listens on a specified port and accepts incoming connections. It ensures that the agent properly handles incoming requests by initiating a listening socket and waiting for the manager to connect, enhancing the system's flexibility in connection management. This adjustment also includes graceful closure of the listening socket. Signed-off-by: SammyOina <sammyoina@gmail.com> --------- Signed-off-by: SammyOina <sammyoina@gmail.com>
Cocos AI
Cocos AI (Confdential Computing System for AI/ML) is a platform for secure multiparty computation (SMPC) based on the Confidential Computing and Trusted Execution Environments (TEEs).
With Cocos AI it becomes possible to run AI/ML workloads on combined datasets from multiple organizations while guaranteeing the privacy and security of the data and the algorithms. Data is always encrypted, protected by hardware secure enclaves (Trusted Execution Environments), attested via secure remote attestation protocols, and invisible to cloud processors or any other 3rd party to which computation is offloaded.
Install
The following prerequisites are needed to run Cocos:
Build Docker Images
Currenty, there is no Docker registry, so Cocos images must be built by hand:
make dockers
For this you might be needing to setup DNS servers in your /etc/resolf.conf as explained
here (i.e. add Google's nameserver 8.8.8.8).
Run Composition
Once the images are built (docker images command should show you ghcr.io/ultravioletrs/cocos/manager),
composition can be run:
make run
This will bring up the Cocos docker services and interconnect them.
Usage
The quickest way to start using Cocos is via the CLI. The latest version can be downloaded from the official releases page.
It can also be built and used from the project's root directory:
make cli
./build/cocos-cli version
Additional details on using the CLI can be found in the CLI documentation.
Documentation
Official documentation is hosted at [Cocos official docs page][docs]. Documentation is auto-generated, checkout the instructions on official docs repository.
License
Cocos AI is a proprietary product created by Ultraviolet company.