CentaurusInfra/alnair

Issues

• GDS traffic monitor

  #147 opened 2 years ago
  0

• [alnair device plugin] feature request -- support GPU selection

  #143 opened 2 years ago
  0

• GPUDirect to local SSD

  #136 opened 2 years ago
  0

• intercept-lib test instruction doesn't work.

  #132 opened 2 years ago
  1

• Add prometheus export to report process-level GPU utilization and memory used size

  #131 opened 2 years ago
  0

• scheduling needs

  #128 opened 2 years ago
  0

• A bad case for dlsym real func acquirement.

  #127 opened 2 years ago
  3

• vGPU scheduler assume all the nodes have GPU information annotation. Cannot handle cpu node or the period before annotation got patched

  #125 opened 2 years ago
  1

• Containerize vGPU server leads cgroup.procs content invisible (leads to process util inquiry always 0, compute control failed)

  #123 opened 2 years ago
  4

• device-plugin installation error, Init:crashloopback

  #122 opened 2 years ago
  1

• vgpu-server container failed to start, "run/nvidia-persistenced/socket" no such device or address

  #119 opened 2 years ago
  0

• profiler add mem-copy-utils from DCGM to reflect application's io requests

  #115 opened 2 years ago
  1

• intercept lib launched through LD_PRELOAD cannot intercept cuda driver API calls with pytorch version >=1.10

  #114 opened 3 years ago
  1

• profiler remove all pod annotation under ai.centaurus.io domain after gpu process is done, which affects scheduler and device plugin

  #113 opened 3 years ago
  1

• use nsight system inside containers

  #112 opened 3 years ago
  1

• Add pre-start hook to all containers in container runtime to support GPU access

  #110 opened 3 years ago
  0

• same node pods communication through unix socket

  #109 opened 3 years ago
  2

• create an exporter to export burst, overuse and window-size metrics to prometheus.

  #108 opened 3 years ago
  0

• setup multiple nodes cluster for kubeshare performance testing

  #107 opened 3 years ago
  1

• setup tf-serving testing environment for kubeshare throughput testing

  #106 opened 3 years ago
  0

• horovod mnist.py has higher utilization number. what does it do?

  #105 opened 3 years ago
  0

• GPU sharing corner case: vGPUs spread to two or more physical GPUs

  #98 opened 3 years ago
  0

• Alnair device plugin find the pod associated with the allocate request

  #95 opened 3 years ago
  0

• Design and Implement a good GPU utilization metrics

  #93 opened 3 years ago
  1

• revise alnair devicepluginserver to connect the running pod/container info with the device

  #92 opened 3 years ago
  1

• Add GPU metrics to Pod metrics for Job metadata

  #91 opened 3 years ago
  0

• Patch Pod Spec Annotations

  #89 opened 3 years ago
  0

• Kubeshare prototyping and compute sharing deep dive

  #88 opened 3 years ago
  4

• how to run the throughput testing on Kubeshare?

  #86 opened 3 years ago
  2

• how to replace kubeshare executables with the local build?

  #85 opened 3 years ago
  2

• what does code-gen.sh do?

  #84 opened 3 years ago
  1

• fairseq multihead_attention, torch.cat cause RuntimeError: CUDA out of memory

  #83 opened 3 years ago
  0

• Kubeshare test cases:

  #80 opened 3 years ago
  8

• Kubeshare is not targeted on a node with multiple gpus?

  #79 opened 3 years ago
  1

• Kubeshare can NOT generate running Pods with Kubectl v1.23.1

  #77 opened 3 years ago
  1

• Serverless for AI

  #76 opened 3 years ago
  0

• Multi-tenant platform security

  #75 opened 3 years ago
  0

• Investigate distributed storage/cache for fast training dataset loading

  #74 opened 3 years ago
  0

• Ziyu's progress update 1/19/2022

  #72 opened 3 years ago
  0

• pytorch dataloader raise bus error: out of shared memory, in kubernetes container

  #71 opened 3 years ago
  1

• how to debug kubeshare and measure the performance?

  #70 opened 3 years ago
  4

• A To-do List for the design of vGPU-Scheduler

  #69 opened 3 years ago
  0

• how can the scheduler track used/free vGPU info?

  #68 opened 3 years ago
  1

• Alnair vGPU compute resource

  #67 opened 3 years ago
  2

• Device plugin, Kubelet (device plugin manager), API server, Scheduler, who knows which GPU is used/free

  #64 opened 3 years ago
  0

• Scheduler extension point data sharing with CycleState

  #63 opened 3 years ago
  1

• Relevant platform trial

  #62 opened 3 years ago
  0

• build docker to add python startup hook

  #61 opened 3 years ago
  1

• explore/implement ml-profiler UI in prometheus

  #60 opened 3 years ago
  0

• custom tensorflow fit or other tensorboard callback to get layer metrics

  #59 opened 3 years ago
  0