[cloud service] there are so many ECS instances on alicloud server, how to select the type? Best practice note

Selection best practices explain

Selection is not about choosing an instance specification , It's about understanding the key features of instance specifications , In stock shortage 、 The product is offline 、 Use preemptive instances and other scenarios , You can have a variety of alternative instance specifications , make the best of Alibaba cloud ECS Flexible features .

We are choosing Aliyun server When it comes to configuration , Most users choose 1 nucleus 2G,2 nucleus 4G,4 nucleus 8G,4 nucleus 16G,8 nucleus 16 These common configurations , Some novice users are making sure CPU And memory configuration , Often do not know how to choose instance specification , It may be helpful to refer to the following optional instance specifications and usage scenarios for common configurations .

at present 618 Alibaba cloud There are also big promotions , This is the best time to buy ！ Click on the link to learn about , Now let's talk about Alibaba cloud What about the technical selection ！

Understand the instance specification family

Start one ECS Before the example , You need to combine performance 、 Price 、 Workload and other factors , Make decisions that are cost-effective and stable . According to the business scenario and vCPU、 Memory 、 network performance 、 Storage throughput and other configuration division , Alibaba cloud ECS Provides a variety of instance specification families , An instance specification family includes multiple instance specifications . The format of instance specification family name is ecs.< Specification family >, The instance specification name is ecs.< Specification family >.<nx>large.

ecs： Cloud server ECS The product code of .
< Specification family >： It consists of small letters and numbers .
- A lowercase letter is an abbreviation of a word , And it marks the performance area of the specification family . The meaning of some lowercase letters is as follows .
  - c： Generally, it means computational （computational）
  - g： Generally speaking, it means universal （general）
  - r： Generally, it means memory type （ram）
  - ne： Generally, it means network enhanced （network enhanced）
- Numbers generally distinguish the release time between the same type of specification family . Larger numbers represent the next generation of specification families , Have a higher cost performance , Low price, good performance .
<nx>large： Represents the instance specification has vCPU Check the number ,n The bigger it is ,vCPU The higher the number of cores .

for example ,ecs.g6.2xlarge For universal g6 An instance specification in a specification family , Have 8 individual vCPU nucleus . Compared with g5、g4、sn2ne Specification family ,g6 It is a new generation of generic instance specification family .

Start the selection

In order to make it convenient for you to compare the performance of the instance when selecting the model , You can learn more about the instance from the following ways .

Instance specification family ： Refer to the documentation for product details of the instance specification family , You don't need an account to log in .
DescribeInstanceTypes： call ECS API, Get the latest performance specifications , But you need to have a registered account .
```
aliyun ecs DescribeInstanceTypes --InstanceTypeFamily ecs.g6
```
Cloud product pricing page ： understand ECS Pricing information for resources 、 The latest offer 、 And a preliminary estimate of the cost of use .
ECS Custom purchase page ： In the instance configuration section of the basic configuration page , Learn more about shopping guide .

Choose according to the usage scenario

The figure below shows Alibaba cloud ECS Some instance specification families and their corresponding business scenarios .

According to the typical application, it is recommended that

If you are using a software or application similar to the one in the figure below , You can select the instance specification family on the right .

Self built service selection recommendation

If you are a self built service , Depending on the application you use , And refer to the selection principle , Select the corresponding instance specification family .

The application type

Common applications

Selection principle

Recommend instance specification family

Load balancing

Nginx

Application features ： Need to support high frequency of new connection operations .

CPU Ability to calculate ： The demand is higher .
Memory ： Not very demanding .

c6、hfc6 series

RPC product

SOFA
Dubbo

Application features ： Network link intensive ; The process runs at a high memory consumption .

g6 series

cache

Redis
Memcache
Solo
CPU Ability to calculate ： Not very demanding .
Memory ： The demand is higher .
Instance specification family ：r6、re6 series
Block storage ：SSD Cloud disk or ESSD Cloud disk

Configuration center

ZooKeeper

There will be a lot of I/O Read and write operations .

CPU Ability to calculate ： Not very demanding .
Memory ： Not very demanding .
Instance specification family ：c6 series
Block storage ：SSD Cloud disk or ESSD Cloud disk

Message queue

Kafka
RabbitMQ

In terms of message integrity , Cloud disk is preferred for storage .

CPU Ability to calculate ： Not very demanding .
Memory and vCPU The ratio is usually 1:1.
Storage ： Not very demanding .
Instance specification family ：c6 series
Block storage ：SSD Cloud disk or ESSD Cloud disk

Container arrangement

Kubernetes

Through the combination of dragon and container , Can maximize the potential of Computing .

ebmc6、ebmg6 series

Big table storage

HBase

Generally, you can choose d series .
If the business exists super high IOPS（Input/Output Operations Per Second） demand , You can choose i series .
d1、d1ne、d2s series
i2 series

database

MySQL
NoSQL
There is a need for elastic expansion of storage , You can choose ECS and ESSD.
about I/O The needs of sensitive businesses , Preference i series .
Example specifications ：c6、g6、r6 series
Block storage ：ESSD Cloud disk
i2 series

SQLServer

because Windows Of I/O Single channel features , Yes I/O The requirement of reading and writing ability is high , Preference ESSD.
ECS The logical and physical sectors of are set to 4K.
Instance specification family ：c6、g6、r6 series
Block storage ：ESSD Cloud disk

Text search

Elasticsearch

Choose memory and vCPU With a bigger one ECS specifications .
Daily need to export database data to ES file , Yes I/O There are requirements for reading and writing .
Example specifications ：g6 series
Block storage ：ESSD Cloud disk
d1、d1ne、d2s series

Real time computing

Flink
Blink

Based on the amount of storage, you can choose ECS General specifications and cloud disk , You can also choose d series .

d1、d1ne、d2s series

Offline computing

Hadoop
HDFS
CDH

Preference d series .

d1、d1ne、d2s series

Common scenarios 、 Game clothes 、 Live video scene recommendation

In this kind of scenario , Performance requirements are expressed as CPU Computationally intensive , You need a relatively balanced ratio of processor and memory resources , Generally selected CPU Match with memory 1:2、 High efficiency cloud disk is selected as the system disk 、 Data disk selection SSD Cloud disk or ESSD Cloud disk . If the business needs stronger network performance , Such as video barrage, etc , You can choose a higher specification instance specification in the same series , Improve the network receiving and contracting ability （PPS）.

Scene classification

Scene segmentation

Recommended specification family

Performance requirements

Processor to memory ratio

General applications

Balanced performance applications , Background applications

g series , Such as g6

Medium dominant frequency , Computationally intensive

1:4

High network receiving and contracting application

g series , Such as g6

High network PPS, Computationally intensive

1:4

High performance computing

c series , Such as c6

High dominant frequency , Computationally intensive

1:2

Game applications

High performance end game

c series , Such as c6

High dominant frequency

1:2

Mobile Games 、 Page to swim

g series , Such as g6

Medium dominant frequency

1:4

Live video

Video forwarding

g series , Such as g6

Medium dominant frequency , Computationally intensive

1:4

Live barrage

g series , Such as g6

High network PPS, Computationally intensive

1:4

Hadoop、Spark、Kafka Big data scenario recommendation

In this kind of scenario , Because it involves different nodes , Performance requirements are more complex , You need to balance the performance of each node , Including calculation 、 Storage throughput 、 Network performance, etc .

The management node ： Treat as a generic scenario , Please see the Common scenarios 、 Game clothes 、 Live video scene recommendation chapter .
Computing node ： Treat as a generic scenario , Please see the Common scenarios 、 Game clothes 、 Live video scene recommendation chapter . Depending on the size of the cluster , Different instance specifications need to be selected . for example 100 You can choose from the following nodes ecs.g6.4xlage,100 More than nodes can be selected ecs.g6.8xlage.
explain Computing node can adopt preemptive instance in billing mode , Achieve cost performance optimization . For more details , Please see the Overview of preemptive examples .
Data nodes ： Requires high storage throughput 、 High network throughput 、 Balanced processor to memory ratio , It is recommended that you use big data type d Series specification family . for example MapReduce/Hive Can choose ecs.d1ne.6xlarge,Spark/Mlib Can choose ecs.d1ne.8xlarge.

database 、 cache 、 Search scene recommendation

In this kind of scenario , The ratio of processor to memory of instance specification is generally higher than 1:4, Part of the software to store I/O Reading and writing ability, timeliness and ductility can be more sensitive , It is recommended that you choose the specification family with high cost performance per unit memory .

Scene classification

Scene segmentation

Recommended specification family

Processor to memory ratio

Data disk

Relational database

High performance , Depending on the application layer, high availability

i series

1:4

Local SSD Storage 、 Efficient cloud disk 、SSD Cloud disk

Small and medium databases

g series , Or other memory proportion is 1:4 Specification family of

1:4

Efficient cloud disk 、SSD Cloud disk

High performance database

r series

1:8

Efficient cloud disk 、SSD Cloud disk

Distributed cache

Memory consumption scenarios in

g series , Or other memory proportion is 1:4 Specification family of

1:4

Efficient cloud disk 、SSD Cloud disk

High memory consumption scenarios

r series

1:8

Efficient cloud disk 、SSD Cloud disk

NoSQL database

High performance , Application layer height is available

i series

1:4

Local SSD Storage 、 Efficient cloud disk 、SSD Cloud disk

Small and medium databases

g series , Or other memory proportion is 1:4 Specification family of

1:4

Efficient cloud disk 、SSD Cloud disk

High performance database

r series

1:8

Efficient cloud disk 、SSD Cloud disk

ElasticSearch

Small cluster , Rely on cloud disk to ensure high availability of data

g series , Or other memory proportion is 1:4 Specification family of

1:4

Efficient cloud disk 、SSD Cloud disk

Big clusters , High availability

d series

1:4

Local SSD Storage 、 Efficient cloud disk 、SSD Cloud disk

Take the database for example , In the traditional way , The business system is directly connected to OLTP database , Data redundancy is mostly through RAID Disk array implementation . choice Alibaba cloud ECS, Your light load 、 All overloaded databases can be deployed flexibly .

Light load database ： Use enterprise class instance specification with cloud disk , Higher cost performance .
Overloading the database ： High storage required IOPS And low read / write latency , It is recommended that you use local SSD type i Series instance specification family （ With high I/O Type local NVMeSSD Local dish ）, To meet the requirements of large and overloaded database .

Deep learning 、 Image processing scene recommendation

In this kind of scenario , Applications need high performance GPU Accelerator , stay GPU and CPU The following suggestions are made on the proportioning .

Deep learning training ：GPU And CPU The recommended ratio is 1:8 To 1:12 Between .
General deep learning ：GPU And CPU The recommended ratio is 1:4 To 1:48 Between .
Image recognition reasoning ：GPU And CPU The recommended ratio is 1:4 To 1:12 Between .
Speech recognition and synthetic reasoning ：GPU And CPU The recommended ratio is 1:16 To 1:48 Between .

Verification and adjustment

When you complete the selection and start using ECS After the instance , It is recommended that you monitor information based on performance over a period of time , Verify that the selected instance specification is appropriate .

Suppose you choose ecs.g6.xlarge, Find instances through monitoring CPU Usage has been low , It is recommended that you log in to the instance to check whether the memory utilization is high , If the memory consumption is high , You can adjust to a more appropriate specification family for processor to memory resource ratio . For more details , Please refer to the following documentation ：

Use cloud server ECS In the process of , If there is a shortage of inventory in the region 、 Instance specification family is off sale 、 Change to a more cost-effective specification family 、 Upgrade configuration, etc , You can change the configuration according to the characteristics of the instance specification family . The following table lists the configuration modification recommendations for some instance specification families , For more details , Please see the Summary of lifting and lowering modes And Introduction of example specification configuration change .

Current instance specification family

Preferred recommendation

Alternative recommendation

sn1、sn2