Placement Groups
- Cluster: single AZ, lowest latency (10 Gbps within group), HPC workloads
- Spread: distinct hardware, max 7 per AZ, high availability
- Partition: grouped on racks, max 7 partitions per AZ, large distributed systems
- Cannot merge groups; instance must be stopped to move
Enhanced Networking
- ENA (Elastic Network Adapter): up to 100 Gbps, default on modern instances
- EFA (Elastic Fabric Adapter): HPC, bypasses kernel for MPI/NCCL, Linux only
- SR-IOV: hardware virtualization for high PPS and low latency
- Supported on most current-gen instances (C5, M5, R5, etc.)
Bandwidth & Throughput
- Instance bandwidth: varies by type (e.g., c5.18xlarge = 25 Gbps)
- NAT Gateway: 100 Gbps burst per gateway, scale with multiple GWs
- Transit Gateway: 50 Gbps per VPC attachment
- VPN: 1.25 Gbps per tunnel, aggregate with ECMP over TGW
Jumbo Frames
- 9001 MTU within VPC (supported on most instance types)
- Traffic leaving VPC: clamped to 1500 MTU
- Direct Connect: supports 9001 MTU (private/transit VIF)
- VPN: max 1500 MTU due to IPSec overhead
Practice Network Optimization Questions
Put your knowledge to the test with practice questions.