This repository contains scripts to aid in understanding the performance of Amazon's Elastic File System (EFS).
Set the following environment variables before proceeding with further steps:.
export AWS_DEFAULT_REGION=...
export EC2_KEY_PAIR=...
export EFS_EVAL_FILE_SYSTEM_ID=...
export EFS_EVAL_STACK_NAME=...
AWS_DEFAULT_REGION should be your preferred region, e.g., us-west-2.
EC2_KEY_PAIR should be the name of the ssh key used to log into EC2.
This should match a KeyName as returned by aws ec2 describe-key-pairs.
EFS_EVAL_FILE_SYSTEM_ID should match the file system identifier of the EFS
under test. It should look something like fs-12594a6d.
EFS_EVAL_STACK_NAME is a name of your choice, e.g., efs-test.
If you are using a non-default profile configure it as:
export AWS_DEFAULT_PROFILE=...
setting AWS_DEFAULT_PROFILE to a the heading of a section within your
~/.aws/credentials file.
We use scripts that control AWS Cloud Formation to configure resources for the experiment.
python3 clusterconfig.py update \
--stack-name=$EFS_EVAL_STACK_NAME \
--ssh-key-name=$EC2_KEY_PAIR \
--efs-file-system-id=$EFS_EVAL_FILE_SYSTEM_ID \
--az-scale-a=4 \
--az-scale-rest=0
You can check on the status of the configuration as follows:
python3 clusterconfig.py status --stack-name=$EFS_EVAL_STACK_NAME
Configure each test instance. This command will install the necessary software.
python3 clusterconfig.py instances \
--stack-name=$EFS_EVAL_STACK_NAME
Generate a list of instances.
python3 clusterconfig.py instanceconfig \
--stack-name=$EFS_EVAL_STACK_NAME \
--efs-file-system-id=$EFS_EVAL_FILE_SYSTEM_ID
When finished you can clean up the resources
python3 clusterconfig.py remove \
--stack-name=$EFS_EVAL_STACK_NAME
This command should be run on the control server
python3 scan_fio.py \
--results /scanres \
--block-sizes=4K,32K,256K \
--modes=randread,randwrite,randrw \
--io-depths=4,8 \
--num-clients=2,4 \
--num-iterations=2 \
--unique-filenames
See results with
python3 analyze.py \
--results /scanres