โ ๏ธ Disclaimer: This tool is provided as-is, without Microsoft support. Odin is an experimental project that accelerates skills and knowledge ramp up for Azure Local, and helps IT architects validate cluster design configurations.
ODIN Sizer for Azure Local
Version 0.22.65 |
Calculate instance hardware requirements based on your workload scenarios and their resiliency and capacity requirements. This tool provides example hardware configurations only. Please consult with your preferred hardware OEM partner for detailed guidance and recommendations on their Azure Local solution offerings.
Workload Scenarios
Add workloads to calculate your Azure Local instance requirements.
START HERE
△
Start by adding a workload
Click one of the workload buttons above to size your Azure Local instance
💻 VMs⎈ AKS Arc🖥 AVD🧠 Foundry Local🔎 Edge RAG🎬 AI Video Indexer💻 GitHub Enterprise Local
Configure your instance topology and storage settings.
Tip: For business or mission-critical workloads, it is recommended to implement two separate Azure Local instances, to enable workload HA/DR capabilities between two locations, or consider a Rack-Aware Cluster deployment type.
We highly recommend using Three-way Mirror for the highest level of resiliency and performance.
Tip: Minimum N+1 capacity must be reserved for Compute and Memory when applying updates (ability to drain a machine). Single Machine instances will always incur workload downtime during updates.
Physical Machine(s) - Example Hardware Configuration
Define the physical hardware specifications for each machine.
โ ๏ธ Selected GPU count exceeds the maximum supported per machine for this GPU model. The count has been capped.
Advanced Settings
Storage Configuration
All-Flash configuration is recommended for increased performance and resiliency.
Disk Configuration (Capacity)
S2D Repair Reservation
Cache Disks
Cache disk type: NVMe
Capacity Disks
Capacity disk type: SSD
S2D Repair Reservation
Hardware Utilization Information
Continue to add additional workloads, if required. Once complete, open the Designer wizard pre-configured with this instance and hardware sizer settings.
Percentage capacity used/available for defined workloads:
Azure Local hyperconverged instance size0 / 16
(includes N+1 machine count, for servicing and redundancy)
Compute (vCPUs) - Consumed0%
0 / 0 vCPUs
Memory - Consumed0%
0 / 0 GB
Usable Storage - Consumed0%
0 / 0 TB
GPU (Physical Units) - Consumed0%
0 / 0 GPUs (across Nโ1 effective machines)
๐ซOne or more resources (Compute, Memory, Storage, or GPU) are at or above 90% utilization. This is not a recommended configuration โ increase per-machine hardware specs, add more machines, or reduce workloads before proceeding to the Designer.
โ ๏ธ
๐ซStorage configuration exceeds Azure Local supported limits. Export is blocked until the configuration is corrected.
Power calculations, verbose information and assumptions
Estimates based on component TDP ratings. Actual power varies by workload and OEM hardware design. Consult your preferred OEM hardware partner for accurate power and rack planning.
💡 Tip: Left-click and hold to rotate ยท Scroll to zoom ยท Right-click and hold to move
Physical Machine(s) ToR Switch BMC Switch Core Router / Firewall Mgmt/Compute SMB Trunk LAG FC Switch SAN Appliance /
Interactive 3D preview of the estimated rack layout. Each server machine occupies 2U, ToR switches occupy 1U each. The cables from each machine to the ToR switches are not shown. Rack-aware deployments show balanced distribution of machines across two cabinets, in real deployments these can be in separate rooms or datacenter locations. โ ๏ธ This is an approximate representation only, contact your preferred hardware OEM partner for detailed physical space requirements for their Azure Local solutions.
Add Workload
Select Azure Region
Choose the Azure region for your Azure Local deployment. This will be set in the Designer wizard.
Autonomous Cloud Endpoint
Enter the Autonomous Cloud FQDN for the ALDO management cluster. This endpoint is used by disconnected operations for the local Azure control plane.
✓ Valid FQDN format
๐ Import Configuration
๐
Import a previously exported ODIN Sizer configuration
Select a .json file that was exported using the Export JSON button. This will restore all workloads, hardware settings, and MANUAL overrides.
โ๏ธ
Import hardware specs from an existing Azure Local machine
Navigate to any machine in your instance in the Azure Portal → JSON View → Copy the full JSON and paste it below. The exact CPU model, core count, and memory will be extracted.
๐Privacy: All processing is performed entirely in your browser. No instance configuration data is stored, transmitted, or sent to any external service. Sanitize any sensitive information before pasting.
๐
Import your VMware inventory from an RVTools export
Select the RVTools .xlsx "all" export. ODIN reads the vInfo sheet (and, if present, vCluster / vHost for context), detects your source clusters, and turns the chosen cluster's VMs into Sizer workloads.
๐Privacy: All processing is performed entirely in your browser. Only aggregate VM counts, VM names, and per-VM CPU / RAM / storage values are read (VM names are used purely to label the workload rows). Hostnames, IP/MAC addresses, OS strings, datacenter / cluster names, datastore names, tags, custom attributes and annotations are never read or stored. If you'd prefer VM names are not read at all, choose the one entry per VM size option. No data is transmitted to any external service.
โ๏ธ Sizer hardware scaling weighting logic
When you add or change workloads, the Sizer auto-picks a machine count, per-machine memory, and a vCPU:core ratio to keep all four resources (CPU / memory / storage / GPU) under 90 % utilisation while a machine is offline for updates (the N+1 rule). The rules it follows, in order:
Prefer more machines over bigger machines. The recommender uses a fixed "preferred" per-machine memory cap of 1.5 TB (or 2 TB for clusters of 10+ machines) when estimating the minimum machine count. Adding a machine is almost always cheaper than jumping a DIMM tier.
Hardware scale-up order: cores โ sockets โ ratio. When demand outgrows the current per-machine hardware, the Sizer first grows physical cores within the selected CPU SKU, then steps up to 2-socket, then loosens the vCPU:core ratio (4:1 โ 6:1 โ 8:1, etc.). Memory is bumped to the next DIMM tier as needed.
Two-pass memory limits.
Conservative pass โ caps per-machine memory at 1.5 TB, then adds machines if utilisation is still โฅ 90 %.
Aggressive pass โ only if the machine loop hit the 16-machine cluster maximum, the Sizer escalates memory up to 4 TB per machine and loosens the vCPU:core ratio further.
Prefer 3 machines over a high-DIMM 2-machine cluster. If a 2-machine solution lands on more than 768 GB per machine, the Sizer auto-bumps to 3 machines. That unlocks a three-way mirror (stronger resiliency โ survives any single machine failure with full data redundancy) and spreads the workload across smaller, cheaper DIMMs.
Auto-upgrade to Disaggregated Storage at 16 machines. If a hyperconverged cluster would need more than 16 machines, the Sizer switches the cluster type to Disaggregated Storage (separate compute + storage racks) rather than packing 3-4 TB DIMMs into a maxed-out HCI cluster.
Step back down when possible. After the aggressive pass, a reduction loop tries to lower the machine count again if the bigger per-machine hardware made it possible โ keeping the cluster as small as the workload allows.
๐ก You are always in control.
Every auto-decision is advisory. If you have a fixed requirement — a specific machine count, a specific memory size per machine, a specific CPU SKU, or a specific vCPU:core ratio — just set it manually. As soon as you change a field, the Sizer locks that choice and the auto-scaler respects it for all future workload changes.
Have a question or feedback? Please raise an Issue in GitHub.
๐ง
Configuration Imported
๐พ
Configure Storage
Set the disk count, disk size, and storage type to match your cluster's physical disks.
๐ฆ
Add Workloads
Add your VMs, AKS Arc clusters, and AVD sessions to see how the existing hardware handles them.
๐
MANUAL Overrides
Imported values are locked as MANUAL โ auto-scaling won't override them.
๐
Review Capacity
Check utilization bars and sizing notes to see if your hardware fits the workloads.
๐
RVTools Import Complete
๐
10% Growth Buffer AppliedAllow for Future Growth was set to 10% to leave headroom on the sized hardware. Change it any time in the configuration panel.
๐พ
Set Your Storage
RVTools reports VMs virtual hard disk usage, not your source cluster's physical storage size. Check the storage configuration, (data disk count, disk size, and storage type (NVMe, SSD)), to ensure these match your current and future workload storage requirements.
๐ข
Estimates, Not a Quote
Imported workloads are derived from the RVTools snapshot (templates excluded; powered-off VMs only if you opted in). If the source VMs requirements change, import updated workload data again, before relying on the cluster sizing examples.
๐
Review Capacity
Check the utilization bars, and the machine processor core count and memory recommendations. Review the Sizing Notes to see how your imported workloads fit into the example cluster hardware configuration.