Cluster overview-Elastic High Performance Computing(E-HPC)-阿里云帮助中心

A cluster is a collection of nodes that runs High Performance Computing (HPC) applications. It provides powerful computing capabilities that a single node cannot. Clusters offer high performance, scalability, stability, and reliability. This topic describes the basic concepts and features of a cluster.

Nodes

Each node in an E-HPC cluster is an ECS instance. Nodes are categorized by function into logon nodes, management nodes, and compute nodes. The following table describes the function of each node type.

Node	Description
Logon node	A node used for remote logon to the cluster. You can perform operations such as debugging, compiling, and installing software, and submitting jobs.
Management node	A node used to manage the cluster. It deploys the scheduler service and the domain account service. Scheduler service: Runs the server-side of scheduling tools such as PBS or Slurm to process job submissions and manage scheduling. Domain account service: Centrally manages user information within the E-HPC cluster. Important The management node is responsible for job management, scheduling, and domain account resolution. Do not compile software, upload or download data, or package data on the management node. This ensures service stability.
Compute node	A node used to execute HPC jobs.

The following table lists the recommended configurations for management nodes and job limits for clusters of different sizes.

Cluster size	Recommended configuration for management nodes	Job recommendations
Number of compute nodes ≤ 100	vCPUs ≥ 16 Memory ≥ 64 GiB	Number of queued jobs < 5,000 Number of unfinished jobs < 10,000
Number of compute nodes ≤ 500	vCPUs ≥ 32 Memory ≥ 128 GiB	Number of queued jobs < 10,000 Number of unfinished jobs < 20,000
Number of compute nodes > 500	vCPUs ≥ 64 Memory ≥ 256 GiB	Number of queued jobs < 10,000 Number of unfinished jobs < 20,000

Images

An image provides the information required for cluster nodes, which are ECS instances. This information includes the operating system and related configuration data. E-HPC supports the following types of images:

Public image: An official image provided by Alibaba Cloud.
Custom image: An image that you create from an ECS instance or a snapshot, or an image that you import.
Shared image: An image shared with you by another Alibaba Cloud account.
Alibaba Cloud Marketplace image: An image provided by a third-party Independent Software Vendor (ISV) through Alibaba Cloud Marketplace.
Community image: An image published by an image provider through the Alibaba Cloud community image platform.
Compute Nest deployment: An image published by a third-party ISV through Alibaba Cloud Compute Nest.

Important

The available images depend on the selected region and zone, the image resources in your Alibaba Cloud account, and the selected instance type for the nodes. The images displayed in the console are those that you can use.
The supported schedulers, domain account services, and shared storage types vary based on the image you select. For more information, see List of schedulers, domain account services, and shared storage supported by images.

For more information, see Image overview.

Schedulers

A scheduler is the software that schedules jobs on the cluster. E-HPC supports the following schedulers:

Type	Scheduler	Displayed in console
PBS	PBS Pro19	pbs19
	PBS Pro18	pbs Note The version of the installed scheduler software depends on the selected image.
	OpenPBS 20
	OpenPBS 22
Slurm	Slurm 22	slurm22
	Slurm 20	slurm20
	Slurm 19	slurm19
	Slurm 17	slurm
GridEngine	Open Grid Scheduler (SGE)	opengridscheduler
Other	Deadline	deadline

Note

The schedulers supported by an image vary. For more information, see List of schedulers, domain account services, and shared storage supported by images.

Domain account services

Domain account services are used to manage cluster users. E-HPC supports the following two domain account services:

NIS: Network Information Service (NIS) is a centralized identity management service. After you create users on the NIS server, you can use these accounts to log on to any new node that joins the NIS domain without creating the same users on each node.
LDAP: Lightweight Directory Access Protocol (LDAP) is used in E-HPC to authenticate user identities and permissions. You can use LDAP to grant permissions to users and add users to groups to create users with different access permissions.

Note

The domain account services supported by an image vary. For more information, see List of schedulers, domain account services, and shared storage supported by images.

Shared storage

User data, scheduler information, and shared job data for an E-HPC cluster are stored in a shared file system. This allows all nodes in the cluster to access the data. E-HPC supports the following types of file systems:

NAS: Includes General-purpose NAS file systems and Extreme NAS file systems.
CPFS: Supports both CPFS-NFS and CPFS-POSIX mount methods.
Other: Non-Alibaba Cloud file storage, such as a self-built NAS storage system.

Note

The shared storage types supported by an image vary. For more information, see List of schedulers, domain account services, and shared storage supported by images.

List of schedulers, domain account services, and shared storage supported by images

The supported schedulers, domain account services, and shared storage types vary based on the image. The following table lists the support for public images.

Note

When you create a cluster, the images, schedulers, and domain account services that are available are those displayed in the console.
In the following table, Custom indicates that E-HPC does not provide the installation. You must install the component yourself.
CentOS 6 and CentOS 8 have reached their end of life (EOL). The Linux community no longer maintains these operating system versions. We recommend that you switch to a different operating system. For more information, see How to switch the repository for CentOS 6 after EOL or How to switch the repository for CentOS 8 after EOL.

Public image	Supported schedulers	Supported domain account services	Supported shared storage
CentOS 7.2 64-bit CentOS 7.3 64-bit CentOS 7.4 64-bit CentOS 7.5 64-bit CentOS 7.6 64-bit CentOS 7.8 64-bit CentOS 7.9 64-bit CentOS 7.9 64-bit UEFI version	PBS Pro18 PBS Pro19 Slurm 17 Slurm 19 Slurm 20 Slurm 22 Open Grid Scheduler (SGE) Deadline	NIS LDAP	General-purpose NAS file system Extreme NAS file system CPFS-NFS CPFS-POSIX
CentOS 8.0 64-bit	Open PBS 20	NIS	General-purpose NAS file system Extreme NAS file system CPFS-NFS
CentOS 6.9 64-bit	PBS Pro18 Deadline	NIS LDAP	General-purpose NAS file system Extreme NAS file system
CentOS 6.10 64-bit	Custom	Custom	General-purpose NAS file system Extreme NAS file system
Alibaba Cloud Linux 2.1903 LTS 64-bit	PBS Pro18	NIS LDAP	General-purpose NAS file system Extreme NAS file system CPFS-NFS
Alibaba Cloud Linux 3.2104 LTS 64-bit	Open Grid Scheduler (SGE)	NIS	General-purpose NAS file system Extreme NAS file system CPFS-NFS
Alibaba Cloud Linux 3.2104 LTS 64-bit ARM version	Open Grid Scheduler (SGE)	NIS	General-purpose NAS file system Extreme NAS file system CPFS-NFS
Ubuntu 20.04 64-bit	Slurm 22	NIS	General-purpose NAS file system Extreme NAS file system CPFS-NFS
Ubuntu 20.04 64-bit ARM version	Slurm 22	NIS	General-purpose NAS file system Extreme NAS file system CPFS-NFS
Windows Server 2022 Windows Server 2019 Windows Server 2016 Windows Server 2012 R2 Windows Server 2008 R2	Custom	Custom	Custom

Users

After you create cluster users, you can submit, debug, and run jobs on the cluster. You can create two types of users with different permissions.

ordinary permission group: Suitable for regular users who only need to submit and debug jobs.
sudo permission group: Suitable for administrators who need to manage the cluster. In addition to submitting and debugging jobs, they can run sudo commands to perform operations such as installing software and restarting nodes.
Important
The root user can be created only during cluster initialization. Do not use the root user to submit jobs. This prevents accidental operations in job scripts from damaging data in the E-HPC cluster.

For more information, see Manage users.

Software

E-HPC provides mainstream scientific computing applications, compiler runtime libraries, and Message Passing Interface (MPI) libraries. You can install and use them as needed. For more information, see Software overview.

Cluster states

Creating: The cluster is being created. This is the initial state during which ECS instances are provisioned.
Installing: Software is being installed on the cluster.
Initializing: The cluster is being initialized. This process includes initializing the root user.
Running: The cluster has been created and is running as expected.
Abnormal: The cluster enters the Abnormal state if the management node is deleted or stopped, or if the scheduler software exits. You can try to repair the cluster. If the repair fails, submit a ticket.
Releasing: The cluster is being released.