Created by FutureBright
AI Solutions on Cisco Infrastructure Essentials | DCAIE : section 3:
AI Network Architectures
POST-TEST section 3
1.
Which network issue can be the most disruptive to AI/ML training jobs if not
properly managed?
low bandwidth
x packet drops
suboptimal load balancing
jitter
Jitter can not only disrupt the data loading timing, which reduces the training
efficiency, but it can also affect the responsiveness of AI/ML models during
inferencing and make the application less reliable.
High latency or packet drops in server cluster communications can prolong or
disrupt training jobs
2.
Which traffic is typically transported over the back-end network?
x RDMA (RoCE)
management
application
vMotion
storage
, Created by FutureBright
front-end network is used for all other communications, such as management,
storage, and user application traffic.
back-end network is used only for GPU-to-GPU connectivity to transport Remote
Direct Memory Access (RDMA) (RDMA over Converged Ethernet [RoCE]) traffic.
3.
Which load-balancing method optimizes AI/ML workloads in an Ethernet
environment?
stateless ECMP
load based on link congestion
packet spray and reorder
round-robin
DSF enables packet reordering to achieve optimal traffic balancing across all the
links, with all flows treated fairly.
4.
What is the role of the back-end network in an AI/ML cluster?
x to connect GPUs across different hosts
to manage user application traffic
to store training data
to provide redundancy for front-end network failures
5.
AI Solutions on Cisco Infrastructure Essentials | DCAIE : section 3:
AI Network Architectures
POST-TEST section 3
1.
Which network issue can be the most disruptive to AI/ML training jobs if not
properly managed?
low bandwidth
x packet drops
suboptimal load balancing
jitter
Jitter can not only disrupt the data loading timing, which reduces the training
efficiency, but it can also affect the responsiveness of AI/ML models during
inferencing and make the application less reliable.
High latency or packet drops in server cluster communications can prolong or
disrupt training jobs
2.
Which traffic is typically transported over the back-end network?
x RDMA (RoCE)
management
application
vMotion
storage
, Created by FutureBright
front-end network is used for all other communications, such as management,
storage, and user application traffic.
back-end network is used only for GPU-to-GPU connectivity to transport Remote
Direct Memory Access (RDMA) (RDMA over Converged Ethernet [RoCE]) traffic.
3.
Which load-balancing method optimizes AI/ML workloads in an Ethernet
environment?
stateless ECMP
load based on link congestion
packet spray and reorder
round-robin
DSF enables packet reordering to achieve optimal traffic balancing across all the
links, with all flows treated fairly.
4.
What is the role of the back-end network in an AI/ML cluster?
x to connect GPUs across different hosts
to manage user application traffic
to store training data
to provide redundancy for front-end network failures
5.
