CloudTik 1.3.1 Release

CloudTik 1.3.1 release include a critical fix:

• The cluster launching failure caused by wrongly checking container home while the container is not started.

CloudTik 1.3.0 Release

CloudTik 1.3.0 release includes the following new features and enhancements:

• Infrastructure and runtimes to support microservices orchestration.
• Service discovery and configurations (Consul)
• Service DNS naming (CoreDNS, Bind, dnsmasq)
• Load balancing - Layer 4 or Layer 7 load balancer working with built-in service discovery (HAProxy, NGINX)
• API Gateway (Kong, APISIX)
• Monitoring (Prometheus, Grafana)
• Database (MySQL, Postgres)

CloudTik 1.2.0 Release

CloudTik 1.2.0 release includes the following new features and enhancements:

• AI modeling for Graph SAGE neural network. Support inductive and transductive learning on homogeneous and heterogeneous graph with single or distributed training.
• AI modeling for transfer learning for common image and text classification PyTorch and TensorFlow models and datasets.
• AI modeling for XGBoost for regression and classification tasks.
• Application: Quick start application for distributed training of common AI models including BERT-large, DLRM, Mask R-CNN, ResNet-50, ResNeXt101, RNN-T, SSD-ResNet-34.
• Application: Disease prediction application using image classification model and text classification model of transfer learning.
• Application: Fraud detection application using Graph SAGE inductive model with XGBoost classification.
• Improve cluster rsync files with docker containers to be robust on avoiding dirty directories.
• Support both cluster local distributed storage and cloud storage for different data storing purposes: cloud storage for the input and final output data; cluster local storage for intermediate data shared in distributed analytics and AI training.
• Improve HDFS Fuse implementation for solving the async release issue (This is solving write after read problem). Improve HDFS NFS implementation to support set file size for setattr. (This is needed when opening and write an existing file in python)

CloudTik 1.1.0 Release

CloudTik 1.1.0 release includes the following new features and enhancements:

• Support managed cloud database for AWS, GCP and Azure. The cluster runtimes can utilize the managed cloud database if it is configured.
• Support virtual provider to allow user to run virtual clustering on a single node providing the similar clustering functionality as public cloud. This is useful for user to develop their applications locally and run it on cloud.
• Support local provider to allow user to run a single cluster with multiple local nodes for making it easy for such cases.
• Enhanced on-premise provider to allow multiple workspaces. Each workspace can run multiple clusters.
• Refactor cluster dump functionality to use exec and rsync based on common abstraction of a provider implementation.
• Refactor runtime infrastructure to provide common and simpler tools for a new runtime implementation.
• Rename ML runtime to AI runtime. Add a prototype for AI modeling (more matured implementation coming soon)
• Many other improvements and bug fixes.

CloudTik 1.0.7 Release

CloudTik 1.0.7 release includes following new features/improvements:

• Support Ray as a new runtime
• Improve ML runtime for oneAPI with MKL and jemalloc
• Support built-in scaling policy for scaling with CPU load or memory load
• Improve health-check command to be able check node health based on health of runtime processes
• Improve process-status command to be able to show processes of running runtimes
• Support resource-metrics command to show real-time cluster resource metrics and resource metrics of cluster nodes.
• Add tools for AI benchmarks using IntelAI models
• Improve the CLI output text with colorful text
• Improve the docker image configure to default the image tag to current CloudTik version if there is no tag specified

CloudTik 1.0.6 Release

CloudTik 1.0.6 is a release for fixes.

• Fix the problems caused by new version of MLflow dependencies.
• Fix the problem of get worker IP with node status.
• Optimized runtime image pulling speed with docker registry in PRC.

CloudTik 1.0.4 Release

CloudTik 1.0.4 major features:

• Support Alibaba Cloud.
• Support Intel Optimizations for PyTorch in oneAPI based ML/DL runtime.

CloudTik 1.0.3 Release

CloudTik 1.0.3 major additions:

• ML Runtime: Horovod with Intel MPI and oneCCL as Horovod communication layer in ML/DL runtime.
• Kubernetes Provider: Support Azure AKS integration for Kubernetes provider. (Azure AKS workload identity is in public preview)
• Many more examples (ResNet50, MNIST) for TensorFlow, Keras, PyTorch and MXNet with various cases.
• Critical fix for docker configuring failure with the latest docker install.
• Critical fix for MLFlow server failure with latest SQLAlchemy package.

CloudTik 1.0.2 Release

CloudTik 1.0.2 major additions:

• Support Fuse for mounting S3, GCS and Azure DataLake cloud storages to local for enabling more ML/DL cases.
• Improved Horovod Gloo and MPI layer to use CloudTik exec as job scheduler to run tasks on the cluster.
• Supported MXNet with separate option and docker image (MXNet conflicts with TensorFlow and PyTorch on protobuf versions)
• Implemented many more examples (ResNet50, MNIST) for TensorFlow, Keras, PyTorch and MXNet with various cases. Integrated TPCx-AI benchmark.

CloudTik 1.0.1 Release

CloudTik 1.0.1 made a few improvements on the machine learning runtime to support MPI.

• Support OpenMPI for Horovod on Spark.
• Support Gloo and OpenMPI for Horovod without Spark