AWS Application Migration Service (MGN) - Copy Your Servers to AWS

AWS Application Migration Service copies your servers to AWS. That's it. It installs an agent on your source machines, continuously replicates data to AWS, and eventually launches copies of your servers as EC2 instances.

Real talk: It's faster than doing it manually, but anyone selling you a "70% reduction in migration time" is probably measuring against someone doing file-by-file copies with SCP. Your mileage will vary dramatically based on how much legacy crap you're dragging along.

What Actually Works Well

The continuous replication approach is solid. While your production servers keep running, MGN quietly syncs changes to AWS. When you're ready to cut over, the downtime is usually measured in minutes, not hours. That part genuinely works as advertised.

For VMware shops, the agentless option via vCenter 6.7+ is a lifesaver if your security team freaks out about installing agents on production boxes. Though honestly, if they're that paranoid about a replication agent, wait until they see what happens when you try to explain egress charges.

OS Support (The Good and Bad News)

AWS MGN supports Windows Server 2016-2022, most modern Linux distributions, but drops Windows 2003 support in 2026

MGN supports most operating systems you actually care about. Windows Server 2016-2022 works fine. Modern Linux distros (RHEL 7+, Ubuntu 18.04+, Amazon Linux 2) are solid.

The gotcha: Windows 2003 support is getting axed in 2026. If you're still running Server 2003 in production in 2025, migration is the least of your problems - that thing should have been put out of its misery years ago.

Network Configuration Hell

Network Requirements: TCP 443 and 1500 to AWS endpoints, plus staging subnet with S3/EC2/IAM access

Here's where things get interesting. MGN needs to talk home during replication, which means opening specific ports (443 and 1500) to AWS endpoints. Your firewall team will want IP ranges. AWS will tell you to use FQDNs. Your firewall team will insist on IPs. AWS will change the IPs. You'll spend a Tuesday morning troubleshooting why replication broke.

Common network failures:

• ECONNREFUSED 443 - firewall blocking AWS endpoints (again)
• EHOSTUNREACH - routing tables fucked up after "minor network changes"
• SSL_ERROR_SYSCALL - corporate proxy intercepting SSL traffic
• Agent log shows Unable to resolve mgn-dr-gateway-1234567890.us-east-1.elb.amazonaws.com - DNS issues

The staging area lives in your AWS account, so you're paying for those instances while replication happens. Budget for it - that t3.micro staging server for your 500GB database is going to cost you about $15/month, plus EBS storage costs. MGN itself doesn't charge per-server fees, but you pay for all the AWS infrastructure during replication.

Integration Reality Check

MGN plugs into Migration Hub for tracking, which is useful if you have dozens of servers to migrate. The dashboard actually works, unlike some AWS consoles. CloudWatch integration gives you replication health metrics, though you'll probably set up alerts after the first time replication silently fails overnight.

Launch templates are better than the old blueprint system, but you'll still need to manually configure security groups, IAM roles, and probably fix whatever broke during the conversion process. Plan for manual cleanup work post-migration.

Useful Resources for Real Implementation:

• AWS MGN Troubleshooting Guide - When shit breaks at 2 AM
• Common Replication Errors - Error codes you'll actually see
• Network Requirements - Firewall rules that matter
• VMware Agentless Setup - When vCenter hates you
• AWS re:Post MGN Discussions - Real user experiences
• MGN Best Practices Blog - Actual implementation tips
• Migration Hub Setup - Progress tracking that works
• CloudWatch Metrics Guide - Monitoring replication health
• Launch Settings Configuration - Getting your instances right
• AWS Direct Connect for Migrations - When bandwidth matters

What You Actually Pay

Storage Costs: GP3 ~$0.08/GB/month, IO2 ~$0.125/GB/month + IOPS charges

MGN gives you 90 days free per server for the service itself - no per-server charges. The gotchas come from paying for all the underlying AWS infrastructure during replication.

The Staging Infrastructure Bill:
Every server you're replicating needs a staging instance running in AWS during the entire process. That t3.small staging server costs about $15/month. Your database server might need an r6i.large for $135/month. Multiply by however many months your "quick migration" actually takes.

Storage Costs Add Up:
EBS storage for replicating a 500GB server runs about $40/month. If you're syncing a 2TB file server, budget $160/month just for storage. And don't forget - you're paying for this storage in your staging area AND your production instance after cutover.

Regional Availability (For What It's Worth)

MGN is available in most AWS regions where you'd actually want to migrate servers. Check the service availability page because AWS doesn't maintain a consistent list anywhere obvious.

The regional thing matters because you can't replicate from us-east-1 to eu-west-1 directly. You'd set up MGN in eu-west-1 and replicate there from your on-premises environment. Plan your data transfer costs accordingly - cross-region transfers aren't free.

Enterprise Features That Actually Matter

MGN Connector - Automates agent deployment if you have VMware vCenter. Saves you from manually installing agents on 100+ VMs. Released in 2024, it works pretty well when your network team hasn't locked down everything.

Wave Management - Groups servers for coordinated migrations. Useful when you need to migrate a three-tier application in a specific order. The wave functionality actually makes sense and works as advertised.

Organizations Integration - If you're migrating across multiple AWS accounts, this helps centralize tracking. More useful for consulting firms managing migrations for multiple clients than single-company migrations.

Performance Reality Check

Performance Metrics: RPO typically 5-15 minutes, cutover downtime under 10 minutes for simple servers

What Works:

• Continuous replication with sub-5-minute RPO when the network is stable
• Cutover downtime typically under 10 minutes for simple servers
• Initial sync performance depends entirely on your Internet bandwidth (budget 24-72 hours for a 500GB server over 100 Mbps)

What Doesn't:

• "Sub-minute RPO" assumes perfect network conditions nobody has
• Replication lag spikes during high I/O periods on source servers
• Network hiccups can cause replication to fall behind for hours

The disaster recovery use case is legitimate - MGN can keep standby copies of your critical servers ready to launch. But calling it "near-zero RTO" is marketing bullshit unless you've automated the entire failover process and tested it monthly.

Essential Cost Management Resources:

• AWS Simple Monthly Calculator - Estimate before you cry
• EC2 Instance Types Guide - Right-sizing your staging instances
• EBS Pricing Details - Storage cost breakdown
• Data Transfer Pricing - Egress charges explained
• MGN Cost Optimization Blog - Hidden cost gotchas
• Direct Connect Pricing - Dedicated bandwidth costs
• CloudWatch Pricing - Monitoring costs
• Regional Service Availability - Check before planning
• AWS Free Tier Details - What's actually free
• AWS Migration Hub Guide - Centralized migration tracking