Scyld Insight is a web based cluster management system with tools for administrators and users. It is another tool that users can access to monitor the cluster and submit jobs. Administrators can configure and provision compute nodes, monitor activity and utilization, and analyze log files.
Scyld Insight is only accessible on campus at https://hpc.mtech.edu:5001/ - if you have an hpc user account, then you can create a Scyld Insight account with the New Account link. We prefer you use the same username, but you select a different if you prefer. After creating the account, you will ask for some information to connect to Beoweb. You can use any display name for the cluster and the check the Beoweb SSL box. The Beoweb/Master Node is hpc.mtech.edu and the Port is 5003. Leave the Beoweb URL empty. In the last box, enter your hpc username and password. The Ganglia Data Source is scyld (on step 2 of 2).
Insight User Tools
The user can customize the different metrics displayed on the Overview page with the Add Block dropdown menu. the Master Node page is reserved for administrators. The Compute Node page shows metrics for the nodes where the user has jobs running. The Workload page provides a graphical user interface (gui) where jobs can be submitted, monitored, and canceled. The Log Viewer can be used to analyze logs within a user's directory by changing the Log Subscription.
Creating an account
To use Insight, you will first need to create an Insight account, which requires an account on hpc.mtech.edu. First select the "Create an account" link and then enter the same username as on hpc and password for the Insight account. After this account is created, there are two steps to link it to your hpc cluster account.
Step 1: Connect to Beoweb
- Cluster Display Name - HPC
- Check the Beoweb SSL box
- Enter Beoweb Master Node hostname: hpc.mtech.edu
- Beoweb Port: 5003
- Leave the Beoweb URL blank
- Enter the previously provided hpc username and passwd
Step 2: Connect to Ganglia
- Select scyld in the dropdown box
- http://hpc.mtech.edu:80 for port and /ganglia/ for URL
Submitting a job
Create a job with the New Job button. Scripts and programs (compiled for linux) can be accessed from a local directory or the master node. Node Resources, like the number of nodes or gpus on a GPU Nodes can be specified in that section. The wall time (maximum real time) can be selected under the Time section.
If Insight is not running after a reboot, then check the scyld-lmx, beoweb and insight services.
More to be added later.