All Products
Search
Document Center

Test the performance of a Super Computing Cluster (SCC) instance

Last Updated: Jun 30, 2021

SCC instances support low-latency networking by using remote direct memory access (RDMA). This meets the parallel computing requirements of Elastic High Performance Computing (E-HPC) clusters. SCC instances have no performance loss in virtualization and can be isolated by virtual private cloud (VPC). You can have direct access to hardware resources. SCC instances are suitable for scenarios such as manufacturing simulation, life science, machine learning, large-scale molecular dynamics (MD) simulation, and weather forecasting. This topic describes how to create an SCC instance and test the performance of the SCC instance.

Background information

SCC provides high-speed RDMA based on ECS Bare Metal Instance to provide high levels of network performance and increase the speedup of large-scale clusters. SCC has all the benefits of ECS Bare Metal Instance and can support high-bandwidth and low-latency networking. For more information, see Overview.

Compared with ECS instances, SCC instances are connected over an RDMA network that has a bandwidth of 50 Gbit/s. Therefore, SCC instances provide much stronger communication capabilities than ECS instances. The following figure shows the port information that is displayed when you access an SCC instance. Eth0 represents the RDMA port and lo represents the VPC port.

Network port

E-HPC supports multiple SCC instance types such as ecs.scchfg6.20xlarge and ecs.scch5s.16xlarge. For more information, see Specifications.

Create an SCC instance

  1. Log on to the E-HPC console.

  2. Create an SCC instance named HPL.test.

    Set the following parameters:
    • Compute Node: In this example, the ecs.scch5s.16xlarge instance type is selected.

    • Software Configurations: Install linpack 2018 and intel-mpi 2018.

    HPL
  3. Create a sudo user named hpltest.

Test the network performance of an SCC instance

  1. In the left-side navigation pane, click Cluster.

  2. On the Cluster page, find the cluster whose network performance you want to test and click Connect.

  3. On the Connect panel, enter the username, logon password, and port number of the instance. Then, click Connect via SSH.

  4. Test the peak bandwidth of the RDMA network based on the following test sample.

    Perform the following steps to test the peak read BPS:

    1. Log on to the compute000 node and run the following command:

      ib_read_bw -a -q 20 --report_gbits   ## Run the command on the compute000 node
    2. Log on to the compute001 node and run the following command:

      ib_read_bw -a -q 20 --report_gbits compute000  ## Run the command on the compute001 node
      Test the peak read BPS

    Perform the following steps to test the peak write BPS:

    1. Log on to the compute000 node and run the following command:

      ib_write_bw -a -q 20 --report_gbits  ## Run the command on the compute000 node
    2. Log on to the compute001 node and run the following command:

      ib_write_bw -a -q 20 --report_gbits compute000  ## Run the command on the compute001 node
      Test the peak write BPS
  5. Test the latency of the RDMA network based on the following test sample.

    Perform the following steps to test the read latency of RDMA network:

    1. Log on to the compute000 node and run the following command:

      ib_read_lat -a   ## Run the command on the compute000 node
    2. Log on to the compute001 node and run the following command:

      ib_read_lat -F -a compute000  ## Run the command on the compute001 node
      Test the read latency of an RDMA network

    Perform the following steps to test the write latency of the RDMA network:

    1. Log on to the compute000 node and run the following command:

      ib_write_lat -a  ## Run the command on the compute000 node
      
      
    2. Log on to the compute001 node and run the following command:

      ib_write_lat -F -a compute000  ## Run the command on the compute001 node
      Test the write latency of a RDMA network

Monitor the bandwidth utilization of an RDMA network

  1. In the left-side navigation pane, click Cluster.

  2. On the Cluster page, find the SCC cluster whose bandwidth utilization you want to monitor and click Connect.

  3. On the Connect panel, enter the name of the root user, logon password, and port number. Then, click Connect via SSH.

  4. Run the following command to monitor the bandwidth utilization of the RDMA network:

    rdma_monitor -s
    root

View the performance data of the nodes in an SCC instance

  1. In the left-side navigation pane, choose Job and Performance Management > E-HPC Tune.

  2. In the Cluster list, select the cluster whose node performance you want to view.

  3. On the NodeView tab, select nodes and metrics to view the performance data of the nodes in the instance.

    View the performance data of the nodes