F3.2 Contain Applications

Question 1

Why are HPC systems applied?

Answer 1

To perform large-scale financial and engineering simulations that demand LOW LATENCY and HIGH THROUGHPUT.

Question 2

HPC workload managers lack micro-service support and deeply integrated container management capabilities. True or False

Answer 2

True

Question 3

A container orchestrator, such as Kubernetes, on its own does not address all the requirements of HPC systems and cannot replace existing workload managers in HPC centers. What can be used?

Answer 3

A hybrid architecture composed of TWO Clusters, an HPC cluster and a cloud cluster, where container orchestration on the HPC cluster can be performed by the container orchestrator (e.g., Kubernetes) located in the cloud cluster can be used

Question 4

What is the role of resource managers, workload managers, or job schedulers in HPC clusters?

Answer 4

They are used to allocate processors and memory on compute nodes to users’ jobs.

Question 5

Name two mainstream workload managers mentioned in the slides.

Answer 5

Slurm and TORQUE

Question 6

In the TORQUE-managed cluster, what components are present on the head node?

Answer 6

A Portable Batch System (pbs) server daemon and a job scheduler daemon

Question 7

What is the advantage of using containers over traditional virtual machines (VMs) in cloud orchestration?

Answer 7

Containers use the dependencies on their host OS, resulting in a more efficient use of resources and faster start-up times.

Question 8

How does Singularity differ from Docker in terms of network configurations?

Answer 8

Singularity does not require additional network configurations, unlike Docker

Question 9

What is Kubernetes based on, and how does it provide services?

Answer 9

Kubernetes is based on a highly modular architecture, and it provides services through ‘deployment,’ specified in YAML files.

Question 10

In the HPC + Cloud architecture, what is the role of Kubernetes and TORQUE?

Answer 10

Jobs are co-scheduled and co-managed by both Kubernetes and TORQUE. Kubernetes handles scheduling on the first level, and TORQUE manages the second level

Question 11

What is the purpose of the login node in the HPC + Cloud architecture?

Answer 11

The login node serves as a bridge between the TORQUE and Kubernetes clusters, submitting TORQUE jobs to the HPC cluster

Question 12

What is TORQUE-operator, and how does it connect Kubernetes and TORQUE?

Answer 12

TORQUE-operator is a tool that connects with Kubernetes by creating a deployment on the Kubernetes cluster. It creates virtual nodes corresponding to TORQUE queues

Question 13

What services are carried out by the Singularity containers in HPC + Cloud architecture?

Answer 13

4 Singularity containers
1. Generate the virtual node,
2. Fetch queue information,
3. Launch TORQUE jobs to the Kubernetes cluster
4. Transfer TORQUE jobs back to the TORQUE cluster

Question 14

What were the two use cases presented from the CYBELE project?

Answer 14

Pilot Wheat Ear and Pilot Soybean Farming

Question 15

What is the goal of the Pilot Soybean Farming use case?

Answer 15

The goal is to use machine learning for soybean farming, developing a prediction algorithm to infer hidden dependencies between input parameters and yield

Question 16

How does FOG overcome limitations of cloud computing?

Answer 16

By utilizing resources close to end devices.

Question 17

Why are containers considered better than VMs for fog computing?

Answer 17

Containers are easily deployable and have high performance, making them preferable in fog computing.

Question 18

What is Kubernetes, and what role does a pod play in it?

Answer 18

Kubernetes is an open-source orchestration platform for container-based applications. A pod is the most fundamental unit in Kubernetes, containing one or more containers

Question 19

How does Kubernetes expose an application to external cluster access, and what is the role of ClusterIP?

Answer 19

Kubernetes exposes an application through a Service, and the service is bound to a ClusterIP, which is a virtual IP address that never changes

Question 20

What is the role of Kube-scheduler in Kubernetes, and how does it select the optimal node for a pod?

Answer 20

Kube-scheduler is the default scheduler in Kubernetes. It watches unscheduled pods, adds them to a waiting list, and selects the best node for the pod based on filtering and scoring steps

Question 21

What are some policies supported by the FILTERING step in Kubernetes scheduling?

Answer 21

Policies supported by the filtering step include
* PodFitsHostPorts,
* PodFitsResources,
* PodFitsHost, and
* CheckNodeCondition

Question 22

Name some policies supported by the SCORING step in Kubernetes scheduling

Answer 22

Policies supported by the scoring step include
* SelectorSpreadPriority,
* BalancedResourceAllocation,
* NodeAffinityPriority
* ImageLocalityPriority

Question 23

What is ElasticFog, and how does it address the resource allocation challenge in a fog computing environment?

Answer 23

ElasticFog is an elastic resource provisioning method for applications on a container-based fog computing platform. It considers network traffic status for resource allocation and dynamically adapts based on real-time traffic information

Question 24

How does ElasticFog use nodeAffinity rules in Kubernetes for pod allocation across locations?

Answer 24

ElasticFog uses the preferred rule of nodeAffinity to allocate pods based on the proportion of incoming network traffic at each location

Question 25

How does ElasticFog demonstrate awareness of changes in network traffic in real-time during performance evaluations?

Answer 25

ElasticFog adjusts the number of pods among locations based on the proportion of requests coming from each location, showing real-time awareness of changes in network traffic status