Resource Allocation Limits and Fair Usage Enforcement

[bogdanfazakas]

Users can currently monopolize node resources by submitting C2D jobs with unrestricted resource allocations, leading to unfair usage and node operator loss of control.

Critical Scenarios:

1. 0 GPU + All CPU: User takes all CPU cores without using GPU
2. 1 CPU + All RAM: User takes all RAM with minimal CPU usage
3. Excessive Resources: No limits on disk or combinations

Impact:

• Single user can block all other users from the node
• Node operators cannot control fair resource distribution
• Inefficient resource utilization
• Poor user experience for queued jobs
• No visibility of limits in Network dashboard

Implement solution suggeste by @alexcos20 bellow

Add optional minimum and maximums per resource

"resources":[
   {
      "id":"cpu",
      "max":1,
      "constraints": [
           {"id":"ram", min: 1, max 3}, // for each cpu you need to have min 1 gb RAM and max of 3GB
           {"id":"disk", min: 10, max 33}, // for each cpu you need to have min 10 gb disc space but no more than 33 GB
     ]
   },
   {
      "id":"ram",
      "max":1
   },
   {
      "id":"disk",
      "max":1
   },
   {
      "id":"gpu",
      "max":4,
      "constraints": [
           {"id":"ram", min: 1, max 3}, // for each gpu you need to have min 1 gb RAM and max of 3GB
           {"id":"disk", min: 10, max 33},
           {"id":"cpu", min: 2, max 4}, // for each gpu you need to have min 2 cpu and max 4 cpu
     ]
   },
]

Pro: flexible
Con: not easy to understand, code complexity (doable)
Everything is optional: you can have constratins for cpu, but not for memory. You can have either min/max or both

[bogdanfazakas]

• add some check that assure users that a job can be run with the remaining issues left unoccupied when starting a job prioritise GPU and can be used

[alexcos20]

Basicly, we have two options

1. have "SKU" - fixed configs, you cannot choose specific resource

"skus": {
 "light": { "cpu":2, "ram":4, "gpu":1},
 "medium": { "cpu":8, "ram":12, "gpu":2}
}

• Pro: easy to understand
• Con: users will overpay. Requires refactor of node, js, cli, dashboard, incentives, etc !!!

2. Add optional minimum and maximums per resource

"resources":[
   {
      "id":"cpu",
      "max":1,
      "constraints": [
           {"id":"ram", min: 1, max 3}, // for each cpu you need to have min 1 gb RAM and max of 3GB
           {"id":"disk", min: 10, max 33}, // for each cpu you need to have min 10 gb disc space but no more than 33 GB
     ]
   },
   {
      "id":"ram",
      "max":1
   },
   {
      "id":"disk",
      "max":1
   },
   {
      "id":"gpu",
      "max":4,
      "constraints": [
           {"id":"ram", min: 1, max 3}, // for each gpu you need to have min 1 gb RAM and max of 3GB
           {"id":"disk", min: 10, max 33},
           {"id":"cpu", min: 2, max 4}, // for each gpu you need to have min 2 cpu and max 4 cpu
     ]
   },
]

• Pro: flexible
• Con: not easy to understand, code complexity (doable)
  Everything is optional: you can have constratins for cpu, but not for memory. You can have either min/max or both