-
(Linux) Ideally, No more Swap
-
The last thing anyone wants to do is troubleshoot performance problems
-
(Preference) It is better for a component to die and be replaced, than to perform slowly
-
Move toward: clustered and autoscaling architectures
-
-
Use (and abuse) disks
-
Configure as many disks as is reasonable
-
Dedicated disk to take advantage of minimum IOPS guarantees
-
-
Use disks to organize data
-
Use disks to separate or combine IOPS performance
-
-
Ditch partitions, use the whole disk
-
Partitions are an obsolete construct.
-
Cloud disks are easily expandable.
-
Snapshots provide easy recovery mechanism
-
See Section 2 (Use and abuse disks)
-
-
Take advantage of startup scripts
-
AWS/GCP user-data/cloud-init
-
Azure - Startup task
-
-
Go ahead, make copies
-
Copy that disk
-
Copy that VM
-
Spin up time: measured in minutes
-
Proposed By
Richard Guo
Notes
Year
2019