A big release focused on security, flexibility and ease of use: quota profiles, switch drift detection and health monitoring, NVIDIA SpectrumX NCP RA support, intelligent VM placement, DPU support, and more.
MetalSoft version 7.4 is now out. This has been a big release, with a lot of improvements. As always, we are responding to the needs of our customers, adding features that improve security, flexibility and ease of use. Here are the highlights.
Security and Access Control
Quota profiles
Quota profiles now replace the previous limits section and provide more flexibility. They are no longer tied to a particular user but instead are tied to a role and an account. Quotas can be set per account and then per user role within the respective account.
Credentials access migrated to permissions
The old credentials access part was moved to native permissions. This allows them to be set via roles.
TACACS support for SONiC switches during ZTP
We now support configuring TACACS during the ZTP process, including management VRF configuration.
Extensions
Extension configurations in the UI
It is now possible to configure certain extension variables in the UI, per site. This removes the need to re-package an extension if any change is required on the configuration.
Ansible execution environment configurable per extension
This enables extension developers to have their own execution environment that packages the extension's dependencies. This replaces the old monolithic execution environment.
Network and Fabric Management
Switch drift detection
Switches are now monitored for drift. When drift is detected (a manual update), an email is sent to the alert list and the significant Kafka queue receives an event.
Switch health monitoring
Switches are now monitored for health, and in case of degradation an alert is sent, and the AI is used to perform an initial investigation.
Network dashboard in the Fabric Manager
A dashboard is now shown in the Fabric Manager displaying an overview of switch health.
Connection detection in the Fabric Manager at the fabric level
It is now possible to detect connections via switches from the UI, removing the need for manual switch-to-switch connection mapping before fabric setup.
NVIDIA SpectrumX NCP RA support
It is now possible to deploy an entire NVIDIA NCP RA underlay and associated overlay networks for east-west GPU-to-GPU traffic. Multiple Scalability Units are supported.
Configlets support
It is now possible to attach certain config snippets to certain provisioning callbacks, allowing more flexibility with network configurations and enabling custom configurations at the interface level, VRF level, etc. The current support is for NVIDIA SpectrumX switches only. Following releases will enable this across all vendors.
Virtualization and AI Infrastructure
Intelligent VM placement
Incus/KVM VMs are now intelligently allocated to specific NUMA nodes relative to GPUs.
DPU support
It is now possible to terminate VLANs on DPUs. The DPU VFs can also be associated with Incus/KVM VMs.
Google Vertex support for AI integration
It is now possible to use Google's Gemini LLM as a backend for ELI via the Vertex AI integration.
Want to go deeper on the ideas behind this release? Here's Alex Bordei, our CPO, explaining the rail-type architectures that power the new NVIDIA SpectrumX support, how intelligent VM placement squeezes more performance out of GPU servers, and why drift detection keeps your switch configurations honest:
Want to see MetalSoft 7.4 in action on your own infrastructure? Book a demo with our team.