Create an E-HPC cluster

Last Updated: Oct 12, 2017

Logon management

Log on to the E-HPC console. If you don’t have an account yet, click Free Account to sign up.

Go to E-HPC -> Clusters, select a region (such as US East 1) and click Create Cluster.

First, please learn about regions and zones.

Note: Except under special circumstances, when creating, managing, or using E-HPC clusters, DO NOT use the ECS console to make any adjustments to individual cluster nodes. Instead, perform all operations on the E-HPC cluster management platform.

Hardware configuration

Zone

In the cluster creation interface that appears, first select the zone. Or you can also choose to have the system allocate a zone by default. To ensure efficient network communication between E-HPC nodes, activate all nodes in the same zone of the same region. For more details, see the explanation of Regions and zones.

VPCs, VSwitches, and security groups

Next, select the VPC to use. Basically, a VPC is an independent and isolated network environment. To use a VPC, you must first create it and its supporting VSwitch. For more information about VPCs, see VPC overview. Click Create VPC or Create Subnet (VSwitch). VPCs and VSwitches can be created together on the same interface. For details, see Create a VPC and VSwitch process (or click Tutorial). Follow the instructions to create the VPC and VSwitch.

Note: When creating VPCs and VSwitches, you must select the same region and zone as the E-HPC cluster.

After creating the VPC, click Refresh and then select the newly created VPC and VSwitch by name from the drop-down menu to be the network components of the E-HPC cluster.

Next, choose whether or not to create a security group. Security groups are used to set network access control for the cluster. If you do not have any special requirements, you can create a security group using the default rules provided by the E-HPC control system. If you want to allow the new E-HPC cluster to communicate with other clusters that already have security groups, select the names of the relevant existing security groups from the drop-down menu. For details, see Security groups.

Payment method and high availability

Currently, the only supported payment method is Pay-As-You-Go. You will be billed based on the duration of activation, in units of hours. This is a post-payment method. The specific price is shown by the final price calculation result at the lower right of the page.

High availability refers to the HA (high availability) function supported by the cluster. This ensures it will not experience SPOF problems. The cluster’s head node and domain account management node both support HA configurations. For example, the PBS cluster head node and the NIS domain account management node both have a master and a slave node. When the master node is not working, the system will automatically switch over to the slave node.

When HA is enabled, the system will automatically add two control nodes.

Node configuration

E-HPC clusters are composed of the following nodes:

  • Control nodes, two mutually-independent nodes
    • Job scheduling node
    • Domain account management node
  • Computing node
  • Logon node

Generally, the job scheduling node only handles job scheduling, while the domain account management node only handles account information. These nodes do not participate in job computation. Therefore, in principle, you can select low-configuration enterprise-level instances (for example, the sn1ne instance with 4 CPU cores) for the control nodes and still ensure high availability. The hardware configuration selected for the computing node is the key aspect that determines cluster performance. The logon node is generally configured as a development environment. It must provide all cluster users with the resources and testing environment needed for software development and debugging. Therefore, we suggest you select an instance with the same configuration as the computing node or with a higher memory ratio. For detailed information on each instance model, see Recommended configurations.

After enabling HA, the system will automatically allocate four control node instances (master/slave scheduling nodes and master/slave domain account management nodes). If you disable HA, it will only allocate two instances (no slave nodes). If you select the GPU Series filter on the instance list, only GPU instance options will be displayed for the Computing Node and Logon Node. You can specify the number of computing node instances to create. By default, only one logon node instance is created.

Shared storage

Next, we will create shared storage. Data for all users, user management, job sharing data, and other information is stored in the shared storage, where it can be accessed by the various cluster nodes. Currently, shared storage is provided by NAS. Moreover, to use NAS, you need supporting mount points and remote directories (see NAS terminology).

Click Create NAS Instance or Create Mount Point. Then, read the instructions in Create a file system and mount point (or click Tutorial) and create a NAS instance and mount point.

Note: When creating a NAS instance, we recommend you select the same region and zone as the E-HPC cluster. When adding a NAS mount point, you must select the same VPC and VSwitch as the E-HPC cluster.

Now, go back to the E-HPC cluster creation interface and click NAS Instance and Mount Point to select the IDs of the newly created NAS instance and mount point. Then, enter the Remote Directory. The resulting mount path is a combination of the mount point and remote directory, specifically: mount point:/remote directory name. For more information, see FAQ: NAS directory mounting rules.

Note: NAS remote directories must be created in advance. If you have no special requirements, you can leave this blank and simply use the NAS root directory.

Now, click Next to go to the software configuration process.

Software configuration

Operating system

Here, we will decide the operating system version of the virtual system image. If you have no special requirements, we recommend the CentOS 7.x series.

Product version

The current E-HPC product version is 1.0.0.

Scheduler and domain account service

Because of the intrinsic functions and features of these software modules, the coexistence of multiple schedulers or multiple domain account management mechanisms will lead to conflicts or even cause data corruption. Therefore, during cluster creation you cannot select more than one scheduler or account management service. We recommend using PBS + NIS.

Other software stacks

E-HPC provides various PaaS platform software, benchmarks, and applications for use in HPC. You can choose to preload these resources based on your actual needs. For instructions on how to use HPC application software, read the Best practices part of the product documentation. The list of application software in these documents may not be complete. The full list of optional software is shown on the interface during cluster creation.

Note: When you choose to preload HPC application software, you must select the dependency software package (such as mpich or openmpi, see the software package name suffix). If you select software with the suffix “-gpu”, make sure that the computing node uses a GPU series instance model. Otherwise, there is a risk that cluster creation will fail or the software may not run properly.

Other basic settings

This includes the cluster name and password settings. Note: Be sure to keep your passwords secure.

Finally, check the configuration inventory and click Confirm to create the E-HPC cluster.

Query creation status

After about 20 minutes, you can go back to the E-HPC cluster control overview interface and view the status of the new cluster.

If all the nodes of the cluster are in the “Normal” status, this indicates that the creation process is complete. Now you can log on to the cluster and perform operations. For operation instructions, see Cluster use.

Thank you! We've received your feedback.