ACA-CloudNative Questions and Answers

1

2

3

4

True

False

Supports continuous training pipelines with composable workflow

Scales out to hundreds of nodes to support AI/ML model training

Supports heterogeneous resource management, including management of GPU, FPGA, and

RDMA-enabled resources

All resources are free

If worker node capacity is more than 1000 cores, consider using EBM (ECS Bare Metal).

You should choose higher instance specifications, because you can run more tasks on a single

worker node this way, improving utilization.

You should choose the lowest possible instance specification for your workers, to reduce costs.

Ensure redundant worker node capacity for each running pod.

Attach data disk to worker nodes, because Docker image, system logs, and temp data will

consume space. Without a data disk, systems can quickly run out of disk space.

Configure Kubernetes standard output in Import Data by Log Service console

Configure where to store logs in each pod

Use PL/SQL Statement from RDBMS

Database lock required (commit and rollback scenario)

Use store procedure/function from RDBMS

Use a Choreographer and Event Bus. In the failed scenario in a service, a Choreographer will use event bus to cancel all the communications among services related.

True

False

True

False

True

False

Log Service (Ingress)

SLB

CMS

Network

Tighly coupled monolith

Microservices

Client server

Agile

Emerging from trial and error

Deployment

DeamonSet

Sidecar

StatefulSet

True

False

True

False

True

False

Domain-Driven Design

Decompose by Business capability

Centralize all components(Authorization, Presentation, Business logic, Database layer) use single-tiered software application combined into a single program from a single platform

Divide and conquer

Decompose by Subdomain

True

False

Cloud Disk

NAS

OSS

CPFS

Local Disk

HDFS