1# VMSS Guide
2 
3Determine when to recommend a Virtual Machine Scale Set (VMSS) over a single VM, and which VMSS configuration to suggest.
4 
5> **Note:** This reference provides quick guidance but may become stale. Always verify VMSS features, limitations, and orchestration mode capabilities by fetching the latest documentation from:
6> - https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/overview
7> - https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-autoscale-overview
8> - https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/orchestration-modes-api-comparison
9 
10## What Is a VM Scale Set?
11 
12A VMSS creates and manages a group of load-balanced, identically configured VM instances. Key capabilities:
13 
14- **Autoscale** — automatically add/remove instances based on metrics or schedules
15- **High availability** — spread instances across fault domains and Availability Zones
16- **Load balancing** — integrate with Azure Load Balancer (L4) or Application Gateway (L7)
17- **Large scale** — up to 1,000 instances per scale set (marketplace images)
18- **No extra cost** — you pay only for the underlying VM instances, storage, and networking
19 
20## When to Recommend VMSS vs Single VM
21 
22| Scenario                                              | Recommend | Reasoning                                   |
23| ----------------------------------------------------- | --------- | ------------------------------------------- |
24| Stateless web/API behind a load balancer              | VMSS      | Homogeneous fleet, autoscale on demand      |
25| Batch or parallel compute jobs                        | VMSS      | Scale out for jobs, scale to zero when idle |
26| Autoscale needed (CPU, queue depth, schedule)         | VMSS      | Built-in autoscale rules                    |
27| Microservices with identical replicas                 | VMSS      | Consistent config, rolling updates          |
28| High availability across zones (many instances)       | VMSS      | Automatic zone distribution                 |
29| Single long-lived server (jumpbox, domain controller) | VM        | No scaling benefit; simpler config          |
30| Unique per-instance configuration                     | VM        | Scale sets assume identical instances       |
31| Quick proof of concept or dev/test                    | VM        | Faster to stand up, lower complexity        |
32 
33## Orchestration Modes
34 
35VMSS supports two orchestration modes. **Flexible** is recommended for all new workloads.
36 
37| Feature                  | Flexible (recommended) | Uniform (legacy) |
38| ------------------------ | ---------------------- | ---------------- |
39| Mix VM sizes in one set  | ✅ Yes | ❌ No |
40| Add existing VMs to set  | ✅ Yes | ❌ No |
41| Availability Zone spread | ✅ Automatic | ✅ Automatic |
42| Fault domain control     | ✅ Yes | ✅ Yes |
43| Max instances            | 1,000 | 1,000 |
44| Spot instances           | ✅ Yes | ✅ Yes |
45| Single-instance VMSS     | ✅ Yes | ❌ No |
46| VM model updates         | Automatic, Manual, Rolling | Automatic, Manual, Rolling |
47 
48> **Warning:** Orchestration mode cannot be changed after creation. Always recommend Flexible unless the user has a specific Uniform requirement.
49 
50## Autoscale Patterns
51 
52| Pattern            | Trigger                                  | Example                                                      |
53| ------------------ | ---------------------------------------- | ------------------------------------------------------------ |
54| **Metric-based**   | CPU, memory, queue length, custom metric | Scale out when avg CPU > 70% for 5 min                       |
55| **Schedule-based** | Time of day, day of week                 | Scale to 10 instances Mon–Fri 8 AM; scale down to 2 at night |
56| **Combined**       | Metric + schedule together               | Baseline schedule with metric burst capacity                 |
57| **Predictive**     | ML-forecasted demand (preview)           | Pre-scale before expected traffic spike                      |
58 
59### Autoscale Best Practices
60 
61- Set a **minimum instance count ≥ 2** for production HA
62- Use a **cool-down period** (default 5 min) to avoid flapping
63- Scale out aggressively, scale in conservatively (asymmetric rules)
64- Monitor with [Azure Monitor autoscale diagnostics](https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-best-practices)
65 
66## Networking
67 
68| Component               | When to Use                                                              |
69| ----------------------- | ------------------------------------------------------------------------ |
70| **Azure Load Balancer** | Layer-4 (TCP/UDP) traffic distribution; most common for backend services |
71| **Application Gateway** | Layer-7 (HTTP/HTTPS) with TLS termination, URL routing, WAF              |
72| **No load balancer**    | Batch/HPC jobs where instances pull work from a queue                    |
73 
74## Cost Estimation Tips
75 
76- VMSS itself is **free** — cost is the sum of per-instance VM pricing
77- Estimate at **min** and **max** instance counts for autoscale budgets
78- Use **Spot instances** in VMSS for up to 90% savings on interruptible workloads
79- Combine with **Reservations** or **Savings Plans** on the baseline instance count
80 
81## Key VMSS Limits
82 
83| Limit                                  | Value                              |
84| -------------------------------------- | ---------------------------------- |
85| Max instances per scale set            | 1,000 (marketplace/gallery images) |
86| Max instances (managed image)          | 600                                |
87| Scale sets per subscription per region | 2,500                              |
88| Scale operations concurrency           | Up to 1,000 VMs in a single batch  |
89 
90## Further Reading
91 
92- [VMSS orchestration modes](https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-orchestration-modes)
93- [Autoscale best practices](https://learn.microsoft.com/en-us/azure/azure-monitor/autoscale/autoscale-best-practices)
94- [VMSS networking](https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-networking)
95- [VMSS Flexible portal quickstart](https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/flexible-virtual-machine-scale-sets-portal)
96