FAQ: Enterprise-level instances
-
What are enterprise-level instances and entry-level instances?
-
What is the fundamental difference between enterprise-level and entry-level instances?
-
Which instances are enterprise-level and which are entry-level?
-
What is the network performance of enterprise-level instances?
-
What limits apply to instance type changes for enterprise-level instances?
-
Can I change the instance type of an entry-level instance to an enterprise-level instance?
FAQ: Persistent memory-optimized instances
FAQ: General-purpose compute-optimized u1 instances
FAQ: Economy e instances
FAQ: ECS Bare Metal Instances
FAQ: SCC
Appendix
What are enterprise-level instances and entry-level instances?
Enterprise-level instances are a category of instance families that Alibaba Cloud launched in September 2016. These instances deliver high performance, stable computing power, and balanced network performance. Because they provide dedicated and stable compute, storage, and network resources, these instance families are ideal for enterprise workloads that require high service stability.
Entry-level instances are a category of instance families designed for small and medium-sized websites or individual users. Compared with enterprise-level instances, entry-level instances emphasize shared resource utilization. As a result, they cannot guarantee stable compute performance, but are more cost-effective.
Enterprise-level vs. entry-level instances
Enterprise-level instances use a fixed CPU scheduling mode. Each vCPU is bound to a physical CPU hyper-thread. Instances do not compete for CPU resources, which ensures stable computing performance backed by a strict Service Level Agreement (SLA).
Entry-level instances use a non-bound CPU scheduling mode. Each vCPU is randomly allocated to any idle CPU hyper-thread. vCPUs from different instances compete for physical CPU resources, which can cause compute performance to fluctuate under high load. Entry-level instances have an availability SLA but no performance SLA.
Enterprise and entry-level instances
Among the available instance families, the entry-level instances are e, t6, t5, s6, n4, mn4, xn4, and e4. All other instance families are enterprise-level instances.
When should I use enterprise-level instances?
For the use cases of different enterprise-level instances, see Instance families and Best practices for instance type selection.
Network performance of enterprise-level instances
The network performance of an enterprise-level instance scales with its instance type. Larger instance types provide higher network performance. For more information about the network performance of different instance types, see Instance families.
Instance type change limits
For more information, see Limits and checks for instance type changes.
Change entry-level to enterprise-level instances
For more information, see Limits and checks for instance type changes.
Direct Redis deployment on persistent memory
Running Redis applications on a persistent memory-optimized instance can significantly reduce the total cost of ownership (TCO) per gibibyte (GiB) of memory. However, to ensure performance, you must modify your Redis applications. The core approach is data tiering: storing hot data in standard memory and cold data in persistent memory.
To minimize the cost of application changes, the re6p instance family provides Redis-specific instance types that allow you to deploy Redis applications by running just a few commands. For more information, see Deploy a Redis application on an instance with persistent memory.
When you purchase the instance, select an instance type whose name is in the ecs.re6p-redis.<nx>large format.
Migrating Redis clusters to persistent memory
During the migration, you must ensure service stability and data reliability. First, purchase a persistent memory-optimized instance, run business services on a small scale, and perform traffic tests to check whether the basic performance and capacity models meet your expectations. If the tests are successful, you can gradually scale out the cluster of persistent memory-optimized instances until the migration is complete.
Direct deployment of parameter servers on persistent memory
Parameter server (PS) nodes store all training parameters of a training cluster. Traditionally, almost all parameters are stored in memory, which requires a large amount of memory and is costly. You can use a persistent memory-optimized instance and configure its persistent memory as memory. Then, you can store all parameters in persistent memory and retain only the hash table in standard memory. This significantly reduces the TCO of the training cluster.
Configure persistent memory as a local disk
You can use a tool to configure persistent memory as a local disk. For more information, see Configure and use persistent memory.
Which applications require high-performance local disks?
To optimize the performance or cost for your I/O-intensive applications, select persistent memory-optimized instances for testing. Persistent memory-optimized instances can resolve some common performance and cost issues. For example:
-
The latency of a single SQL query is high, or a more stable response time (RT) is required.
-
Resource loading is slow in application scenarios such as game frontends, high-load databases, and high-load web applications.
-
Purchasing large-capacity disks for higher IOPS or bandwidth while leaving the extra capacity idle results in unnecessary costs.
In typical I/O-intensive application scenarios, you can use persistent memory-optimized instances to balance performance and costs. Examples:
Redis and other NoSQL databases such as Cassandra and MongoDB
Structured databases such as MySQL
I/O-intensive applications such as e-commerce, online games, and media applications
Search scenarios that use solutions such as Elasticsearch
Live video streaming, instant messaging, and room-based online games that require persistent connections
High-performance relational databases and OLTP systems
Direct deployment of Redis and MySQL on persistent memory disks
Application modification is not required because Redis and MySQL applications recognize the persistent memory as a standard SSD.
Persistent memory disk performance comparison
You can run a test by performing the following steps:
-
Configure persistent memory to be used as a local disk and mount the disk.
For more information, see Configure and use persistent memory.
-
Use a disk performance testing tool to test the disk performance.
For information about how to use Fio to test disk performance on a Linux system, see the cloud disk performance test commands in Test the performance of a block storage device.
How reliable is persistent memory?
The reliability of data in persistent memory depends on the reliability of the physical server and the persistent memory device. This creates a risk of a single point of failure. Implement data redundancy at the application layer and store long-term business data on cloud disks to ensure data reliability.
In addition, when you release a persistent memory-optimized instance, the data in persistent memory is automatically cleared. Back up your data before you release the instance. Releasing a persistent memory-optimized instance takes longer than releasing a regular instance because of the time required to clear data from persistent memory.
u1 instance use cases
u1 instances are intended for small and medium-sized enterprises. They are suitable for most general-purpose scenarios that do not have high requirements for CPU computing power and performance, such as web applications and websites, enterprise office applications, offline data analytics, and small and medium-sized databases.
Unsuitable use cases for u1 instances
-
Application scenarios that rely on high computing power and performance to deliver a superior user experience, such as gaming and high-frequency trading. Use c7, g7, or r7 instances for these scenarios.
-
Application scenarios that have high requirements for performance consistency, such as high availability (HA) scenarios where the primary and secondary servers must have the same performance. u1 instances support multiple server platforms, and the business performance may vary significantly among the platforms. Use c7, g7, or r7 instances for these scenarios.
u1 instance processor models
u1 instances support Intel® Xeon® Platinum 6149 (Skylake), 8163 (Skylake), 8269CY (Cascade Lake), 8260 (Cascade Lake), 8369HB (Cooper Lake), and 8369HC (Cooper Lake) processors.
Specifying processor models
u1 instances do not support specifying a processor model.
u1 processor frequencies
u1 instances support server platforms that have various levels of computing power. The CPU base and turbo frequencies of a u1 instance depend on the CPU model of the host on which the instance is located. The actual base and turbo frequencies may change because the instance may be migrated to different CPU platforms during its lifecycle.
Migration of u1 instances across server platforms
During its lifecycle, a u1 instance may be migrated to a new host due to reasons such as underlying host upgrades or O&M. The new host may use a different processor from the original host. All types of instances may be migrated during their lifecycles. For more information about possible processor changes after a u1 instance is migrated, see General-purpose compute-optimized instances (u1).
Unsuitable use cases for e instances
Enterprise-level business scenarios that require high performance stability and a committed performance SLA for cloud servers.
e instance CPU models and specification
e instances use Intel® Xeon® Platinum scalable processors and do not support specifying a processor model.
e instances are built on mainstream, cost-effective CPU platforms. They may be migrated to different CPU platforms during their lifecycle. Technical measures ensure business compatibility across different platforms.
e instance processor frequencies
The base frequency is 2.5 GHz. Because e instances may be migrated to different CPU platforms during their lifecycle, the turbo frequency is not guaranteed.
Migration of e instances across server platforms
During its lifecycle, an economy e instance may be migrated to a new host due to reasons such as underlying host upgrades or O&M. The new host may use a different processor from the original host.
The underlying host is managed by Alibaba Cloud. This type of instance relies on Alibaba Cloud's resource pooling technology and intelligent scheduling algorithms to perform dynamic resource management. This provides sustained computing power, stability, supply, and elasticity for your applications.
Comparing ECS Bare Metal, VMs, and physical servers
For more information about the differences among ECS Bare Metal Instances, traditional cloud servers (virtual machines), and traditional physical servers, see ECS Bare Metal Instance types.
Network performance of ECS Bare Metal Instances
The network performance of an ECS Bare Metal Instance scales with its instance type. Larger instance types provide higher network performance. For more information about the network performance of different instance types, see Instance families.
Supported disk types and disk attachment
ECS Bare Metal Instances support ESSDs, standard SSDs, and ultra disks. You can attach a maximum of 16 data disks to an ECS Bare Metal Instance.
Instance type changes and failover support
ECS Bare Metal Instances do not support instance type changes. When an ECS Bare Metal Instance encounters a hardware failure, failover is supported. All data is retained on cloud disks.
Logical processor count mismatch in Windows 10 Pro
Some ECS Bare Metal Instance types have more than two physical CPU chips, but Windows 10 Pro supports only a maximum of two CPUs. This means only two physical CPU chips can be used. To resolve this issue, switch to a Windows Server public image provided by Alibaba Cloud.
Using the SCC RDMA feature
When you create an SCC instance, select an SCC-customized public image. This image supports the Remote Direct Memory Access over Converged Ethernet (RoCE) driver and the OpenFabrics Enterprise Distribution (OFED) stack. You can use the RDMA feature through InfiniBand (IB) verbs programming or use Message Passing Interface (MPI) for RDMA communication.
Procedure: On the Public Image tab, select Alibaba Cloud Linux for the operating system, select 3.2104 LTS 64-bit SCC Edition for the version, and then select Security Hardening.