Storage reads are slow, although disks are healthy. What is worth checking?
Why D is the answer
Network path, mount options and client tuning
Storage performance also depends on the network and client.
In detail
- Storage speed also depends on the network and client.
- Performance depends on the client too.
Use case
An illustrative scenario, not a real incident.
Disks are healthy, but a mount option limits reads.
See it in different ways
mindmap
root((Verification))
Compute
HPL
ClusterKit
Fabric
NCCL
Signal quality
Versions
Switches
BlueField
Transceivers
Storage
fiomindmap
root(("Storage path"))
n1["Check"]
n2["Network path"]
n3["Mount options"]
n4["Client tuning"]Key point: Check network, mounts and client.
Why the other options are wrong
- A. The GPU fan curve in the BMC
- A fan curve is not in the storage path.
- B. The sampling temperature of the model
- Temperature is a model setting.
- C. The GPU power limit on each node
- A GPU power limit is not in the storage path.