Compute Performance: EC2 (Elastic Compute Cloud) Options#

“Which instance type / placement / purchase option” questions test whether you know the instance family alphabet and the three placement group shapes. (Pricing is covered in pricing-models.)

Instance families#

Family	Optimized for	Mnemonic
T (t3, t4g)	Burstable general purpose (baseline CPU + credits)	“t” = tiny/test
M (m5, m6i)	Balanced general purpose	“m” = middle
C (c5, c7g)	Compute intensive (app servers, game servers, batch)	“c” = compute
R (r5, x2idn) / X / Z	Memory intensive (in-memory DB, caches, big data)	“r” = RAM
P / G / VT / Inf	GPU/ML training and inference	“p/g” = graphics
Hpc / D	High performance compute, high disk throughput
Mac	macOS build farms

Burstable (T family) detail: instances earn CPU credits when idle and spend them under load; “unlimited” mode can pay for extra credits. “Variable low CPU baseline” → T family.

Placement groups - the three shapes#

Type	What it gives	Limit / use
Cluster	Instances in one AZ (Availability Zone), same rack → lowest latency, highest throughput	HPC, tightly-coupled, low-latency networking
Spread	Instances on distinct hardware (max 7 per group per AZ)	Critical instances that must not fail together
Partition	Hardware partitions (up to 100s of instances)	Big data: HDFS, HBase, Cassandra, Kafka

Exam keywords: “lowest possible network latency between nodes” → cluster; “critical instances must not share hardware” → spread; “HDFS / Kafka / topology-aware” → partition.

flowchart TB
    subgraph cluster[Cluster placement: one AZ, one rack]
        a[EC2] --- b[EC2] --- c[EC2]
    end
    subgraph spread[Spread placement: distinct hardware]
        d[EC2] .. e[EC2] .. f[EC2]
    end

Networking and storage performance#

ENA (Elastic Network Adapter) (enhanced networking): 25 Gbps+ , required for modern types; EFA (Elastic Fabric Adapter) = OS-bypass for HPC/ML
Instance store = raw NVMe on the host → highest IOPS (input/output operations per second), but ephemeral (data dies with the instance)
EBS (Elastic Block Store) gp3/io2 = network block storage; io2 for provisioned IOPS (databases); st/sc1 for cheap cold
AMI (Amazon Machine Image) choice: pick the right region/architecture (see creating-and-using-your-own-ami)

Exam check#

Q1: A tightly-coupled HPC simulation needs the lowest possible latency between 20 nodes. Which placement group?

A) Spread
B) Cluster
C) Partition
D) Multi-AZ group

Cevabı göster

Answer: B. Cluster placement = single rack, single AZ, minimum latency.

Q2: Three critical instances must run on separate physical hardware so one host failure can’t take them all down. Which placement?

A) Cluster
B) Partition
C) Spread
D) Auto Scaling

Cevabı göster

Answer: C. Spread = distinct hardware (7 per AZ limit).

Ozet#

T = burstable (CPU kredisi), M = dengeli, C = hesaplama, R/X = bellek, P/G = GPU
Cluster placement = tek AZ, en dusuk gecikme (HPC)
Spread = ayri fiziksel donanim, maksimum 7/AZ (kritik instance)
Partition = donanim bolmeleri (HDFS, Kafka, Cassandra)
Instance store = en hizli disk ama gecici; EBS = kalici ag diski
ENA = hizli ag, EFA = HPC/ML icin bypass