The aarna.ml GPU Cloud Management Software (CMS) addresses the needs of GPUaaS providers (Neoclouds, NVIDIA Cloud Partners or NCPs, sovereign cloud providers, and private cloud providers) by offering self-service on-demand multi-tenancy. Through our software, our customers can offer bare-metal-as-a-service, virtual-machine-as-a-service, Kubernetes clusters with GPUs, job-submission-as-a-service, and model-as-a-service to their tenant users.
Exciting New Features in CMS 5.1
We are thrilled to announce the latest release of CMS, version 5.1! It brings a host of powerful new features designed to enhance your experience and streamline your workflows.
With CMS 5.1, GPUaaS providers can accelerate revenue growth by offering differentiated, high-value services like bare metal, VM, Kubernetes clusters, and model serving—all with seamless self-service experiences that attract more enterprise and AI/ML customers. Enhanced multi-tenancy, advanced networking, and security features enable providers to serve diverse workloads securely, increasing their pricing power and appeal to regulated industries. Deep integrations with billing, payments, and usage-based pricing unlock new monetization models, ensuring every GPU cycle is billable while reducing revenue leakage. At the same time, operational costs are slashed through automation of provisioning, deprovisioning, and quota management, freeing up teams from manual tasks. Overall, CMS 5.1 helps providers grow faster, improve margins, and deliver enterprise-grade services with minimal overhead.
Let's look at what's new.
CMS Deployment support
- CMS is now supported on ARM architecture along with x86.
- CMS is now supported on Red Hat OpenShift Container Platform (OCP).
Platform Management Enhancements
- Integrate platform features (GPU allocation, job submission) with Tenant User Personas, so that single pane-of-glass can be used for all Personas (Admin/Tenant Admin/Tenant User)
- Integration with HuggingFace, so that the users can pick models and run jobs or use them for Inferencing
- Model Serving for Inferencing, with a Play Ground for testing the Inference capabilities with ability to tweak parameters
Networking Enhancements
VPC, Security Groups, and Gateways
- Support for Virtual Private Cloud (VPC)
- Support for Subnets with overlapping IPs
- Network isolation at both L2 & L3
- Configure Security Groups for granular access control
- Configuring Border Leaf & Gateways for External network connectivity
- Integration with our partner F5 Networks for External Gateway
Enhancements to InfiniBand (IB) Multi-Tenancy
- Integration with UFM for creating per tenant compute and storage networks.
CaaS/Kubernetes Enhancements
- NCP tenant users can create multinode K8s clusters, in a fully isolated manner, using bare metal or VM instances
- Support for Upstream Kubernetes distribution, with Openshift support coming soon!
Bare Metal-as-a-Service Enhancements
- Enables NCP tenant users, to provision bare metal servers, using either NVIDIA BCM (Mission Control), in addition to RedFish/Canonical MaaS based support that was already existing
- Additional support for Metal post provisioning automation
- GPU workload deployment
- Disk erase on de-allocation
VM-as-a-Service Enhancements
- Several functional enhancements to VMaaS (based on KVM Hypervisor)
- Fully isolated, Secure & high performance VMs with GPU, NIC, NVMe and MACVTAP pass through
- Volume erase on de-allocation
Integration with Billing and Payments Software
- Ability for the Super-Admin Persona to create Services (that include compute, storage, network traffic etc.), and set pricing
- Tight integration with our partner Billing solution from Monetize360 (M360), providing a single pane of glass for Admin and Tenant Personas
- The Billing solution provides the ability to define Contracts with the Tenants, generate invoices, view and make payments and many more!
Security and RBAC
- Federated identity (OIDC/SAML) with multi-tenant RBAC
- MFA / SSO for the Tenants
- JWT propagation & validation across microservices.
- CMS platform split mode deployment, where Front end microservices can be deployed in different networks/DC/Cloud, and the Back end microservice deployment on nodes labelled according to the network they access.
- Fine-grained authorization using OPA/Gatekeeper policies is coming soon!
Tenant Management Enhancements
Default Tenant
- Establish a Default Tenant that initially owns all resources
- Allow resource reservation by other tenants
- Implement mechanism to transfer resource ownership from the Default Tenant, with option to set aside minimum resource commitment
Tenant Quotas
- Enable Global Admins to set and modify quotas for tenants
- Provide Tenant Admins the ability to request quota adjustments
Integrations
BCM Integration
- Integrate with NVIDIA Base Command Manager (BCM) for Bare Metal (BM) provisioning
Storage Integration
- Integrate with VAST & DDN storage controllers, in addition to existing support for Vanilla NFS, Lustre & S3 support
- Support for Parallel file systems reachable over RoCE, NFSoRDMA
NVCF Integration1
- Integrate with NVIDIA Cloud Foundation (NVCF) for Kubernetes (k8s) cluster management
- Enable manual and CLI-driven scaling of k8s clusters
OpenShift Integration for NVCF
- Integrate OpenShift for NVCF Input/Output Handling (IOH)
- Support OpenShift CRC for development and testing
- Provide OpenShift CRC setup instructions
NVCF Billing
- Implement NVCF billing based on actual GPU utilization metrics
- Ensure billing reflects usage, not allocation, to the NVCF tenant
User Management and Personas
Tenant Admin/User Personas
- Define distinct Tenant Admin and User Personas
- Support multiple Identity Providers (IdProviders)
- Integrate with LDAP and Keycloak
New Features Summary
New Features Summary
Interested in learning more? You can:
- Set up a meeting with us
- Try aarna.ml GPU CMS for free on a digital twin environment without any real hardware
- Try aarna.ml GPU CMS for free on your hardware
Contact us at info@aarna.ml for any of the above.
