Phase Verbs / Action Phrases
| Phrase | Meaning |
|---|---|
| place (instances) | control which physical hardware EC2 instances land on |
| isolate (from failure) | ensure one hardware failure doesn’t affect another instance |
| span (multiple AZs) | distribute instances across different availability zones |
| distribute (data/servers) | spread workload across partitions for fault tolerance |
| group (together) | cluster instances on adjacent hardware for low latency |
Technical Vocabulary
| Term | Definition |
|---|---|
| Placement Group | A logical grouping of EC2 instances that influences their physical placement on AWS hardware |
| Cluster | Placement strategy that packs instances on same hardware in one AZ for max network speed |
| Spread | Placement strategy that puts each instance on separate hardware; max 7 per AZ |
| Partition | Placement strategy using rack-level isolation; up to 7 partitions per AZ; scales to hundreds |
| Rack | A physical unit of server hardware in a data center |
| Enhanced Networking | EC2 feature providing high bandwidth (10 Gbps) low-latency networking |
| HDFS | Hadoop Distributed File System — big data storage that spans many nodes |
| HBase | Distributed NoSQL database built on HDFS |
| Partition-aware | Application designed to know which partition/rack its data lives on |
| PTP (Precision Time Protocol) | High-accuracy hardware clock sync protocol |
| NTP (Network Time Protocol) | Standard internet time sync (less precise than PTP) |
Key Concepts
Why Placement Groups?
- Control how EC2 instances are physically placed relative to each other
- No direct hardware control — you tell AWS your preference, it honors it
4 Placement Group Strategies
| Strategy | AZ Scope | Max Per AZ | Key Benefit | Key Risk |
|---|---|---|---|---|
| Cluster | Single AZ only | Unlimited | 10 Gbps bandwidth, ultra-low latency | Single AZ failure = all down |
| Spread | Multiple AZs | 7 per AZ | Each instance on separate hardware | Limited scale (7/AZ) |
| Partition | Multiple AZs | 7 partitions, hundreds of instances | Rack-level isolation | Partitions can still affect each other in same AZ |
| Precision Time | Any | — | Direct access to high-precision time source | Only for time-sensitive apps |
Cluster Placement Group
- All instances in one AZ, on adjacent hardware
- Get ~10 Gbps network between instances
- Pro: Ultra-fast networking, low latency
- Con: If AZ goes down, everything fails simultaneously
- Use for: Big data jobs, HPC, anything needing extremely low latency + high throughput
Spread Placement Group
- Each instance on a different hardware rack
- Can span multiple AZs
- Limit: Max 7 instances per AZ per placement group
- Pro: Maximizes instance isolation; failure of one hardware ≠ failure of others
- Con: Hard limit means small groups only
- Use for: Critical applications where simultaneous failures must be prevented
Partition Placement Group
- Instances grouped into partitions; each partition = a different rack
- Up to 7 partitions per AZ; can span multiple AZs in a region
- Can hold hundreds of EC2 instances total
- Partition-level failure is isolated — partition 1 failure ≠ partition 2 failure
- Application can query the instance metadata service to know which partition it’s on
- Use for: Distributed big data workloads — HDFS, HBase, Cassandra, Apache Kafka
Precision Time Placement Group
- Instances placed on hardware with a direct connection to high-precision time sources
- Clock syncs to Amazon Time Sync Service at hardware level (much more precise than NTP)
- Has access to PTP hardware clock and hardware packet timestamping
- Use for: Distributed databases, financial timestamping, strict event ordering
Exam Tips
- Cluster = performance (low latency, high throughput); single AZ; high risk
- Spread = availability (isolated hardware); max 7 instances per AZ
- Partition = scalable isolation (rack-level); up to hundreds of instances; Hadoop/Cassandra/Kafka use cases
- Spread ≠ Partition — spread is per-instance, partition is per-rack with many instances per partition
- Precision Time = niche; only when sub-microsecond time accuracy is required