Increase i3 instance store restore performance by 2.5x

Monday, Jun 3, 2019| Tags: ebs, ec2, instance store, aws, performance, IO, st1

EC2 i3 instances are great for scenarios where you require very high read/write performance as it provides you with locally attached instance store volumes upto 2 million read IOPS and 1.6 million write IOPS (for i3en.24xlarge). You can find details on available performance on these volumes at . Many AWS customers use these instances for high IOPS sensitive applictions like databases.

However instance store volumes are ephemeral. In order to ensure durability you need to be able to push the data on local SSD volumes to EBS on a regular basis. Throughput optimized HDD (st1) volumes ae a popular choice for this use case ( In the event of data loss due to loss of i3 instance you will need to restore data from your EBS volume to a new i3 instance store SSD.

Loading of this data from an st1 volume could be slow hampering RTO. One of myc ustomers was looking to reduce the time to restore.

Reading for st1 volumes for restoring to instance store is a read heavy operation and can benefit from increasing the read-ahead setting. You can find more details on how to do it at

I ran a test and was able to improve the performance by ~2.5 times. Look at the result of the test that I performed.

Source Target Size (GB) Time taken (sec) Time taken (min) Throughput (MB/Sec) Read Ahead configuration
st1 ssd 1000 8025 133.75 128 sudo blockdev –report /dev/nvme10n1
RO RA SSZ BSZ StartSec Size Device
rw 256 512 512 0 10737418240000 /dev/nvme10n1
st1 ssd 1000 3297 54.95 311 sudo blockdev –setra 2048 /dev/nvme10n1
sudo blockdev –report /dev/nvme10n1
RO RA SSZ BSZ StartSec Size Device
rw 2048 512 512 0 10737418240000 /dev/nvme10n1