On-demand disk pattern

The on-demand disk pattern is similar in nature to the scale up pattern, in that it operates on an already running instance and requires some downtime. This pattern is a manual process and might be avoided in automated processes such as the scale out pattern, as the downtime involved might cause alarm or scaling policies to trigger, causing some unwanted side effects.

The benefit of on-demand disk size is that you do not need to plan ahead for disk resources. Once the instance is running, you can simply resize its volume when it gets to its maximum capacity. Another example would be in the event where your application is very I/O heavy. At first your instance performs very well, but over time, with increased usage, the I/O operations take longer. An easy way to get more performance out of the base volume might be to enable RAID and stripe the volume.

First we will launch the AWS Linux AMI with an additional 20 GB EBS volume. Ensure that the Volume Type is Magnetic and it is set to Delete on Termination as shown in the following screenshot:

Once it is running, we will need to format and mount the drive for usage. We will mount it to /opt/data:

Now that we have the EBS volume mounted to /opt/data, we will write a file to the disk to ensure that a resize retains the data:

Volume resize

First we will unmount the volume from the running instance with the umount command:

After the volume is unmounted, select the instance from the console, and select the 20 GiB drive attached as /dev/sdb. This will take you to the corresponding volume in the Volumes section of the console. Click the Actions drop-down, and select Create Snapshot, as described in the snapshot pattern. Once this snapshot is 100 percent available for use, detach the /dev/sdb volume from the instance as shown in the following screenshot:

To increase the size of the volume, select the snapshot, and select Create Volume from the Actions menu. Here, select a larger size such as 30 GB, and click Create. When the new volume is available, click Attach Volume from the Actions menu, and select the instance that we detached, ensuring you attach it back to the device we detached it from, or in this case /dev/sdb. Back inside the virtual machine, you will remount the device, check it for errors, resize the partition, check the new size, and ensure that we didn't lose any data:

Change volume from magnetic to SSD

One method to get some increased I/O performance is to change the volume from magnetic to a general purpose SSD, which has a throughput of about 128 MB/s. To do this, select the snapshot that we created for the volume, and click Create Volume from the Actions drop-down menu. Once this volume is available, attach it as /dev/sdb to the instance. From the running instance we will mount it back, ensure that no loss of data has occurred, and check for an increase in throughput:

Here, we can see that we nearly doubled our throughput.

Increase I/O through software RAID

Tip

If you wish to read more about the different software RAID levels and how they perform in the AWS virtualized environment, more information can be found at their User Guide at http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/raid-config.html.

As seen, the throughput for the new file was about 1.6 GB/s for creation and about 40 MB/s for a rewrite. Magnetic EBS volumes are estimated between 40 and 90 MB/s. This is a very rough...