Difference between revisions of "JIRIAF Meeting Nov. 2 2023"
Jump to navigation
Jump to search
Line 47: | Line 47: | ||
##### JRM image and accessibility | ##### JRM image and accessibility | ||
### Prototyping virtual kubelete-based k8s cluster | ### Prototyping virtual kubelete-based k8s cluster | ||
+ | ### Running multiple user workflows within a single JRM | ||
+ | #### Resource assignment and isolation | ||
+ | #### User workflows (PODs) are shell commands. How do they communicate back to the k8s App server? | ||
+ | ## '''M2''' | ||
+ | ### Develop alternative mechanisms for monitoring user workflow. | ||
+ | #### Evaluate Prometheus (complete) | ||
+ | #### Develop and prototype Prometheus exporter within the JRM | ||
+ | ##### Set of hardware metrics that can differentiate between various processes operating within a single user container and across multiple containers. | ||
+ | ### Prometheus server at the JCS | ||
+ | #### Develop Prometheus scraper for monitored hardware metrics | ||
+ | #### Develop algorithms defining state and performances for deployed user workflows. | ||
+ | ## '''M3''' | ||
+ | ### Define mechanisms to act on user workflows, such as | ||
+ | #### Reduce previously allocated resources to the user workflow/application | ||
+ | #### Stopping user application | ||
+ | #### Using workflow-specific control mechanisms to manage it. | ||
+ | ##### This assumes storing application control metadata as part of the user job request. | ||
+ | ### Define mechanisms to control JRM | ||
+ | #### Stop JRM when requested wall time is elapsed | ||
+ | #### Stop JRM when all processes (pods) within the JRM are completed. | ||
+ | ### Report back and update the ''available resources'' table with remaining wall-time released resources by completed processes. | ||
+ | ### Define and reinforce inactivity timeout, after which JRP will be terminated. | ||
+ | |||
+ | |||
Revision as of 14:25, 2 November 2023
Connection Info:
You can connect using the following link (Meeting ID: 161 690 3130). (Click "Expand" to the right for details -->):
One tap mobile: US: +16692545252,,1608518798# or +16468287666,,1608518798#
Meeting URL: https://jlab-org.zoomgov.com/j/1616903130?pwd=cjg3U0Y4SndXL05SeFBmQjVHZkhrQT09&from=addon
Meeting ID: 161 690 3130
Passcode: 018094
Join by Telephone
For higher quality, dial a number based on your current location.
Dial:
US: +1 669 254 5252 or +1 646 828 7666 or +1 551 285 1373 or +1 669 216 1590 or 833 568 8864 (Toll Free)
Meeting ID: 161 690 3130
International numbers
Join by SIP
1616903130@sip.zoomgov.com
Join by H.323
161.199.138.10 (US West)
161.199.136.10 (US East)
Meeting ID: 160 851 8798
Passcode: 018094
Agenda:
- Announcements
- Summary of the project's undertakings and key achievements
- M1
- Examine the potential for incorporating Kubernetes (k8s) into the JCS. (Complete)
- JRM development and prototyping
- VK-cmd
- Running VK-cmd using SLURM
- Running VK-cmd using SWIF2
- SSH tunneling between JCS/k8s APP server and remote JRM
- JRM image and accessibility
- VK-cmd
- Prototyping virtual kubelete-based k8s cluster
- Running multiple user workflows within a single JRM
- Resource assignment and isolation
- User workflows (PODs) are shell commands. How do they communicate back to the k8s App server?
- M2
- Develop alternative mechanisms for monitoring user workflow.
- Evaluate Prometheus (complete)
- Develop and prototype Prometheus exporter within the JRM
- Set of hardware metrics that can differentiate between various processes operating within a single user container and across multiple containers.
- Prometheus server at the JCS
- Develop Prometheus scraper for monitored hardware metrics
- Develop algorithms defining state and performances for deployed user workflows.
- Develop alternative mechanisms for monitoring user workflow.
- M3
- Define mechanisms to act on user workflows, such as
- Reduce previously allocated resources to the user workflow/application
- Stopping user application
- Using workflow-specific control mechanisms to manage it.
- This assumes storing application control metadata as part of the user job request.
- Define mechanisms to control JRM
- Stop JRM when requested wall time is elapsed
- Stop JRM when all processes (pods) within the JRM are completed.
- Report back and update the available resources table with remaining wall-time released resources by completed processes.
- Define and reinforce inactivity timeout, after which JRP will be terminated.
- Define mechanisms to act on user workflows, such as
- M1
- AOT
Useful References