Okay, here is the abstract and bulleted summary for the provided TrueNAS Scale transcript:
Abstract:
This video revisits the virtualization capabilities within TrueNAS Scale after a two-year period, evaluating its current state as a potential all-in-one storage and VM server host. Following a hardware recap involving overcoming HBA connectivity issues in a high-performance NAS build, the creator demonstrates the VM creation process and tests performance and stability. Key findings include strong compute performance but significant issues with storage latency on spinning disk pools. The management interface and functionality, specifically the Virtual Console (Spice/VNC) and GPU passthrough, are found to have multiple bugs and inconsistencies, including unresponsive inputs, plain text password prompts, and unstable GPU passthrough requiring host reboots. The conclusion is that TrueNAS Scale's virtualization stack has not progressed significantly in two years and remains unsuitable for anything beyond basic use cases, falling short when compared to dedicated hypervisors.
Evaluating TrueNAS Scale Virtualization in 2024
0:00 Introduction: Revisiting TrueNAS Scale virtualization after two years to assess improvements and determine its viability as a hypervisor.
1:32 Project Recap: The project involves a high-performance desktop NAS build (45 Drives HL8, Ryzen 9 7945HX CPU, 64GB RAM) intended as a fast VM host.
2:29 Hardware Challenge: The Mini-ITX motherboard lacked sufficient SATA/SAS ports, requiring an HBA. An initial attempt with a fragile M.2 card failed.
3:44 Hardware Solution: The issue was resolved by using two Silverstone ECS07 5-port M.2 SATA HBAs, providing sufficient ports (10 total, 8 for data, 2 for boot drives).
5:14 TrueNAS Host Approach: The goal is an all-in-one server. The chosen approach for this evaluation is running TrueNAS Scale as the host OS and virtualizing VMs directly on it, allowing ZFS greater access to server memory.
6:02 Previous Evaluation Context: Two years prior, TrueNAS Scale virtualization was noted as promising but lacking features like memory ballooning and polish in the interface compared to dedicated hypervisors.
6:36 Current Evaluation Goal: To see if TrueNAS Scale is now ready to compete with established hypervisors after two years of development.
6:49 VM Setup Prerequisite: Requires a configured storage pool. The creator used a 7-disk RAID Z2 array but recommends NVMe drives for VM storage due to later findings.
7:21 VM Creation Wizard: Walks through the multi-step VM creation process in the TrueNAS Scale UI, covering basic info, CPU/memory allocation, disk, networking, and optional GPU passthrough.
8:31 Memory Ballooning Added: Dynamic memory allocation (ballooning) has been added since the previous review, allowing VMs to use memory from a larger pool as needed. Caution is advised against over-provisioning.
9:38 VM Compute Performance: Basic VM functionality works, and compute performance (tested with Cinebench R15 in a VM) is very strong, demonstrating the host CPU's power.
10:10 Storage Performance Issue: A major disappointment is poor random read/write storage performance for VMs stored on spinning disk ZFS arrays (10ms average latency vs 0.03ms for NVMe), making the VM feel unresponsive.
Virtual Console (Spice/VNC): Often stops responding to mouse/keyboard inputs if not the active window, requiring closing and reopening the console.
Password Entry: The Virtual Console password prompt displays the password in plain text, deemed a significant oversight.
Password Inconsistency: When adding a virtual display manually after VM creation (due to a wizard bug), the password is limited to 8 characters, unlike the wizard which allows longer passwords.
11:54 GPU Passthrough Bugs: The process is broken via the initial VM creation wizard. Isolating the GPU must be done beforehand. Even when configured correctly via the devices screen, passthrough was unstable, leading to VM crashes (e.g., with an Intel A310 in Ubuntu) and requiring a TrueNAS host reboot to release the GPU.
14:40 Overall Conclusion: The creator expresses disappointment, finding little significant progress in TrueNAS Scale's virtualization stack over two years. Usability issues and bugs persist, some potentially representing a regression. Basic issues like lack of advanced networking (VLANs) also remain.
16:14 Recommendation: TrueNAS Scale's virtualization is not recommended for anything beyond the most basic use cases. Other dedicated hypervisors are suggested as superior alternatives.
16:29 Comparison to Proxmox: For an all-in-one server, the creator still recommends running Proxmox as the host OS and virtualizing TrueNAS within it, despite some limitations (e.g., RAM allocation for ZFS), finding this approach more stable and manageable overall when virtualization is needed.
17:59 Final Takeaway: An unstable hypervisor that is difficult to manage is not a good solution for most users.