Google Cloud Account Number: Compute Engine General Model Selection Guide!
In the architectural design of Google Cloud (GCP),
General Purpose (General-Purpose)
It's like the "universal toolbox" of cloud computing ". Whether it is building an enterprise official website, running a micro-service container, running a small and medium-sized database, or doing daily development tests, more than 80% of the business will eventually fall on general-purpose models.
But open the GCP console and you'll find that the general-purpose genealogy is densely packed.
E2, N1, N2, N2D, N4, C3, C4, Tau T2D, Axion (N4A/C4A)
A series of code names. If you choose the wrong one, you will either spend more money or your business will be stuck during the peak period.
This article will use the most down-to-earth "actual combat language" to dismantle the naming code and selection logic of GCP general-purpose models, and take you to choose the one with the highest cost performance.
1. first understand the naming rules: GCP's "letter password"
GCP's general-purpose model naming may seem complicated, but it follows a clear formula:
$$\ text {Family Classification} + \text {Generational Number} + \text {Chip Vendor/Architecture Suffix }$$
1. Series classification (prefix letter)
E (Cost-Optimized / Efficient): Extreme price/performance ratio. Suitable for lightweight loads that are not delay sensitive, require smooth computing power, or allow CPU sharing/bursting.
N (Balanced / Network & Compute): Equilibrium Flagship. The CPU and memory ratio standard (usually 1 vCPU : 4GB RAM) provides smooth and pollution-free computing power and is the first choice for the main business of enterprises.
C (High-Performance General Purpose): High-performance general-purpose type. Although it belongs to the general-purpose family, it is equipped with the latest high-frequency CPU and Google Titanium chip architecture to deal with heavy-duty applications with high concurrency, high IO/storage throughput and high throughput requirements.
2. Chip suffix (architecture identification)
No suffix: Intel architecture (e. g. N2, N4, C3, C4). The best compatibility, enterprise old system migration insurance.
D (AMD):AMD EPYC architecture (e. g. N2D, N4D, C3D, C4D). The number of cores and memory bandwidth are large, usually 10%-20% cheaper than Intel models of the same generation.
A (Arm / Google Axion):Arm architecture (e. g. T2A, N4A, C4A). Based on the self-developed Axion chip (Arm Neoverse core), it can extract amazing energy consumption and price performance ratio in cloud native container and microservice scenarios.
2. three main camps dismantling: who should you choose?
In order to let you make quick decisions, we divide the general-purpose models into three major
Camp:
cost reduction artifact, steady main force, performance monster
.
------------------------------------------------------------------------------------------
│ GCP Universal Selection Decision Tree │
└────────────────────┬────────────────────┘
│
----------------------------------------------------------------------------
▼
[Development/Test/Low Frequency] [Production Environment/High Availability]]
│ │
▼
What is the demand bias for the E2 series?
(Even shared core)
----------------------------------------------------------------------------
▼
[Pursuit of Extreme Cost Performance] [Pursuit of High Concurrency/High Frequency/Delay Sensitivity]]
│ │
Is the architecture compatible?
│ Select C series (C3/C4)
-------------------(Titanium hardware acceleration)
▼
[Arm architecture] [x86 architecture]]
│ │
▼
Select N4A / T2A Select N4 / N2D
Camp 1: Extreme Cost Reduction-E2 Series
Representative Model
:
e2-micro
,
e2-small
,
e2-medium
,
e2-standard-*
Low-level features: dynamic scheduling (pooling) based on physical machine CPU resources. Among them, the shared core (Shared-Core) model allows the CPU to burst for a short time (Bursting), but usually only occupies part of the physical core.
Advantage: Extremely cheap! e2-micro also enjoy a free monthly quota in the designated area.
Pit Avoidance Guide: Never Use E2 Shared Core to Run High and Concurrent Production Database! Because when multiple tenants preempt the CPU at the same time, the CPU resources will be flattened, resulting in serious delay jitter (CPU Steal).
Best Scenarios: Internal dev/test environments, blogs/lightweight websites, DevOps CI/CD build nodes, automated scripting tools.
Camp 2: Enterprise Steady Main Force-N Series (N2 / N2D / N4 / N4A)
If you're not sure how much performance your business needs, you can never go wrong with the N-Series.
Model Series
Chip Platform
Key Highlights
Recommended Application Scenarios
N2
Intel Ice Lake / Cascade Lake
Traditional standard general-purpose type, supports custom CPU/memory ratio (Custom Shapes)
Traditional three-tier architecture Web, small and medium MySQL/PostgreSQL
N2D
AMD EPYC
It is 10%-15% cheaper than N2, supports up to 224 vCPU, and is suitable for large concurrency scenarios.
Internal middle-end system, batch task, container cluster
N4 (Latest)
Intel 5th Generation Emerald Rapids + Titanium
The new king of cost performance, introducing dynamic resource management, significantly improved performance compared to N2
In this large scale
Web services, high I/O databases
N4A (Axion)
Google Self-developed Arm (Neoverse N3)
The physical core has no hyperthreading (1 vCPU = 1 physical core), and the cost performance is greatly reduced.
Go/Node.js/Java(Arm) Containerized Microservices
Focus on N4 and N4A:
Launched by GCP
N4
The series marks the overall trend of general-purpose models.
Titanium Architecture
(Google's Offload acceleration chip, which strips network and storage processing from the main CPU). If your business is running on N2, it is recommended to gradually evaluate the migration
N4
, you can usually get higher and smoother throughput with lower prices.
However, if your technology stack has been containerized (K8s / GKE), try it.
N4A (Axion Arm)
. Since there is no need to consider virtual hyper-threading interference (each vCPU is a real physical core), it has very bright performance and cost advantages when dealing with API interfaces and microservices.
Camp 3: High Performance General Purpose-C Series (C3 / C4 / C4D)
Many people will be confused: "isn't C Compute-Optimized (computational optimization)?"
In the latest classification of GCP,
C3/C4 is actually classified as a high-performance branch of the general-purpose family.
.
Core advantages: high frequency and the latest instruction set: equipped with Intel 4th/5th Gen Xeon or AMD EPYC Turin processor. DDR5 memory and Titanium offload: network throughput can reach up to 200 Gbps, and disk IOPS is greatly improved. Bare Metal and Local SSD: suitable for high-performance scenarios without virtualization overhead.
Application scenarios: Redis/Memcached cache with high concurrency and high QPS, MySQL/PostgreSQL production library with heavy read and write, real-time advertisement bidding system (Ad Tech), and high-load game server.
The golden rule of 3. actual combat selection (recommended collection)
Faced with so many options, architects can follow the following four golden steps when making architecture selection decisions:
1. AMD (suffix D) or Arm (suffix A) option is preferred
Unless your software is bound with an Intel-exclusive instruction set (such as a specific AVX-512 encryption algorithm or an old compiled binary package):
Choting N2D / C4D can usually directly save 10% ~ 15% of the bill expenditure compared with the pure Intel version.
Selecting N4A / C4A can take advantage of the high energy of Arm architecture
significantly reduce the computational cost per unit of QPS.
2. Distinguish between "CPU sharing" and "exclusive core"
Internal development, testing, CI/CD nodes: select E2 (even e2-micro / small shared core).
Any online API and database with SLA commitment: must start with N4 / N2 / C4 exclusive core and never apply shared core.
3. Clever use of Custom Machine Types (custom specifications)
A big killer of the GCP N series is
Custom CPU and Memory Combinations
.
If your application consumes a lot of memory but does not consume much CPU (for example, a Python / Java background requires 16GB of memory, but 2 vCPUs are enough), you do not need to buy predefined ones.
n2-standard-4
(4vCPU/16GB), directly create a
2 vCPU + 16GB RAM
to avoid paying for unused vCPUs.
4. Combine Spot VM with CUD (promise to use discount)
Stateless web/container nodes: Turn on Spot VM (preemptible instances) and get an amazing discount of 60%-90%.
Core database and basic services: sign a 1-year or 3-year CUD (Committed Use Discounts) and lock in a long-term discount of 30%-55%.
4. Summary: A Table Seat
Business Scenario
Recommended Selection
Reasons for selection
Personal Blog/Test Environment/CI Tools
E2 (Shared Core)
Extremely low cost, supporting short CPU bursts
Daily Enterprise Web Application/Small and Medium Database
N4 / N2D
Smooth computing power, excellent cost performance, support for custom configuration
Cloud Native Container Cluster (GKE / Docker)
N4A (Axion Arm)
Physical core without hyper-threading interference, high concurrent processing ability, high energy efficiency
High concurrency database/real-time bidding/game service
C4 / C3D
Titanium hardware accelerated network and storage, high frequency, low latency
The essence of cloud computing selection is in.
Upper Performance Limit
with
Monthly Bill
Find the best balance. I hope this guide will help you clear the fog on your route and choose the GCP computing engine that best suits your current business stage!