Alibaba ECS Instance Types Comparison

6 min readApr 26, 2019

Alibaba Cloud offers a number of ECS instance types optimized to meet various needs at the enterprise level. Instance types are specialized for different purposes, and vary by virtual CPU (vCPU), disk capability, memory size, and other features, offering a number of options to match any workload.

ECS instance types at the enterprise level are available for computing on the x86-architecture or for heterogeneous computing, with generous options among both. But with so much to choose from, how does one find the right ECS instance type to meet their individual needs? We drew up a comparison for a quick guide on the different options and what they offer. The chart below and written descriptions are a brief and easy reference, but remember that finding the right instance type for your workload will always depend on your needs.

General Purpose

All general purpose ECS instance types are I/O optimized, offer a CPU to memory ratio of 1:4 and come with support for SSD and Ultra Cloud Disks.

g5

The g5 instance type is backed by a 2.5 GHz Intel Xeon Platinum 8163 processor. With an ultra high packet forwarding rate, higher computing specs and higher network performance, this type is ideal for scenarios involving the transfer of a large volume of packets, running enterprise-level apps, small to medium database systems, data analysis, and computing clusters and data processing that rely on memory.

sn2e

Similar to the g5, the sn2e instance is ideal for the same scenarios and has higher computing specifications and enhanced network performance. Backed by a 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) or Platinum 8163 (Skylake) processors.

hfg5 (high clock speed)

The hfg5 is centered upon stable performance. Having many features in common with the g5 and sn2e options, this type family also offers a 56 vCPU instance type for boosted performance. With a 3.1 GHz Intel Xeon Gold 6149 (Skylake) processor, this instance type is better suited for high-performance front end servers, science and engineering apps, and Massively Multiplayer Online (MMO) games and video coding.

Compute Optimized

Compute optimized ECS instance types are all I/O optimized with support for SSD Cloud Disks and Ultra Cloud Disks, but vary in their vCPU to memory ratios and ideal scenario uses. Read on to learn the nuances.

Se1ne (enhanced network performance)

The se1ne comes with higher computing specs for enhanced network performance, a vCPU ratio of 1:8, ultra high packet transfer rate for receiving or transmitting, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) or Platinum 8163 (Skylake) processors. This type is ideal for large pack transfers, high-performance and large memory databases, data analysis, mining, distributed memory cache, and enterprise-level apps with large memory requirements (Hadoop, Spark).

hfc5 (high clock speed)

Optimized for stable performances, the hfc5 type boasts 3.1 GHz Intel Xeon Gold 6149 (Skylake) processors, vCPU to memory ratio of 1:2, higher computing specs and performance, and ideal for high-performance Web front-end servers, science and engineering applications, and MMO gaming and video coding.

gn6v, gn5, gn5i, gn4, ga1 (with GPU)

The g-series ECS instance type families come are all optimized for GPU compute workloads, with varying GPU processors and compute to memory options. All are ideal for deep learning, scientific computing, high-performance computing, rendering, multi-media coding, decoding, and other GPU compute workloads.

gnv6: V100 GPU processors, vCPU to memory ratio of 1:4, 2.5 GHz Intel Xeon Platinum 8163 (Skylake) processors.
gn5: NVIDIA P100 GPU processors, no fixed ratio of vCPU to memory, high-performance local NVMe SSD disks, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors.
gn51: NVIDIA P4 GPU processors, vCPU to memory ratio of 1:4, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors.
gn4: NVIDIA M40 GPU processors, no fixed CPU to memory, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors
ga1: AMD S7150 GPU processors, vCPU to memory ratio of 1:2.5, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors, and high-performance local NVMe SSD disks. Also ideal for other server-end business scenarios that require powerful concurrent floating-point compute capabilities

f1, f2, f3 (with FPGA)

The f-series ECS instance type families are ideal for deep learning, genomics research, financial analysis, picture, transcoding, and computational workloads, including real-time video processing and security.

f1: Intel ARRIA 10 GX 1150 FPGA, vCPU to memory ratio of 1:7.5, 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors
f2: Xilinx Kintex UltraScale XCKU115, vCPU to memory ratio of 1:7.5, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors
F3: Xilinx 16nm Virtex UltraScale + VU9P, vCPU to memory ratio = 1:4, and 2.5 GHz Intel Xeon Platinum 8163 (Skylake) processors

Memory Optimized

Memory optimized ECS instance types are optimized to meet for needs for high-performance and high memory databases and come I/O optimized with support for SSD and Ultra Cloud disks.

re4

The re4 is ideal for memory-intensive applications and Big Data processing engines (Apache spark or Presto), with 2.2 GHz Intel Xeon E7 8880 v4 (Broadwell) processors, up to 2.4 GHz Turbo Boot, vCPU to memory ratio of 1:12, up to 1920.0 GiB memory, and ecs.re4.20xlarge and ecs.re4.40xlarge have been certified by SAP HANA.

se1ne, se1

The s-series instance types are good for transfers of large volumes of packets, data analysis and mining, distributed memory cache, and Hadoop, Spark, and other enterprise-level applications with large memory requirements.

se1ne: vCPU to memory ratio of 1:8, high packet transfer rate, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) or Platinum 8163 (Skylake) processors.
se1: vCPU to memory ratio of 1:8, 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors

Big Data

d1ne, d1

The d-series ECS instance types are I/O optimized with support for SSD and Ultra Cloud disks come with high computing specs and network performance. They’re best used for Hadoop MapReduce, HDFS, Hive, HBase, etc, Spark in-memory computing, big data computing and storage analysis (i.e. internet and finance industries), Elasticsearch, logs, and so on.

d1ne: High-volume local SATA HDD disks with high I/O throughput and up to 35 Gbit/s of bandwidth for a single instance, vCPU to memory ratio of 1:4, 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors
D1: High-volume local SATA HDD disks with high I/O throughput and up to 17 Gbit/s of bandwidth for a single instance, vCPU to memory ratio of 1:4, and 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors

Intensive Compute

ic5

The ic5 instance types stands on its own with a vCPU to memory ratio of 1:1. I/O optimized with support for SSD Cloud Disks and Ultra Cloud Disks, a high packet transfer rate, and 2.5 GHz Intel Xeon Platinum 8163 (Skylake) processors. Ideal for web front-end servers, data analysis, batch compute, and MMO game front-ends.

With Local SSD

i2, i2g, i1

The i-series type family all come with high-performance local SSD disks with high IOPs, high I/O throughput, and low latency. They’re ideal for OLTP and high-performance relational databases, NoSQL databases (Cassandra and MongoDB), and search applications, such as Elasticsearch

i2: vCPU to memory ratio of 1:8, designed for high-performance databases, 2.5 GHz Intel Xeon Platinum 8163 (Skylake) processors
i2g: vCPU to memory ratio of 1:4, designed for high-performance databases, 2.5 GHz Intel Xeon Platinum 8163 (Skylake) processors
i1: vCPU to memory ratio of 1:4, designed for big data scenarios, 2.5 GHz Intel Xeon E5–2682 v4 (Broadwell) processors

What Alibaba ECS instance types are right for your workloads?

With a wide range of ECS instance types belonging to multiple families, how do you determine the ECS type is a good match for your workload? With plenty to choose from, it’s highly probable that one of them will meet your needs effectively, but first you need to know what those needs are. Once you have a clear and ongoing vision of your workload, usage trends, and business needs, use the guide to start looking for the ECS type that’s right for you.

Further reading:

Originally published at www.parkmycloud.com on November 8, 2018.