Cloud-Aided Yocto Build Speedup
Cloud services have been a reality for some years already. Right now, one of the biggest concepts in the field is the Internet of Things, or IoT for short. When we electronics folk think of the cloud, the first thing that comes to our minds is connecting several devices to it and uploading data to be visualized, stored and analyzed - among other possibilities. Of course, other uses may come to mind, too: cloud services can also be used for storage and processing power, particularly in software-as-a-service applications and personal usage; IT professionals and sysadmins may take advantage of cloud-based VMs and databases.
One thing we don't always realize is that the cloud can be much more than simply an IoT tool, or something only large corporations have access to. We can make use of the cloud for everyday tasks, too. As a matter of fact, that's the motivation behind this article - taking a time-consuming task and using the cloud to simplify our lives a little!
Since I'm familiar with Azure, which is Microsoft's cloud platform, I will show you how to provision an Ubuntu Server 16.04 LTS virtual machine (yes, a Linux machine on Microsoft's cloud!) in Azure, and use Yocto/OpenEmbedded to generate an embedded Linux image, just like the standard image available for the Toradex Colibri iMX6 computer on module.
The first step is to go to portal.azure.com and sign in to Azure with your Microsoft account.
After doing that, simply follow the instructions on this page from Azure's documentation to quickly create a virtual machine.
Tip #1: In your computer's terminal, after generating an authentication key using the following command:
ssh-keygen -t rsa -b 2048
you will need to copy the public authentication key. Just copy the contents of the “id_rsa.pub” file. Execute the command below and copy its output:
Tip #2: After completing the “Connect to Virtual Machine” step, you may stop. You do not need to install NGINX as described in the Azure article.
By accessing the VM via ssh, you can prepare the build environment. To do so, follow the instructions in this technical article on Toradex's developer site.
Tip #1: Despite the article's length, all you need are these sections: “Prerequisites” - Ubuntu 16.04, “Installation – Setup/Configure”, “Installation – local.conf” - to select Colibri iMX6, and “Building”.
Tip #2: When the VM is initialized, it will have a 30 GB disk (SSD) on which the operating system is installed. There is also a volatile disk in the /mnt folder, and its size varies according to the VM configuration you choose.
Note: you can increase the size of the 30 GB disk (SSD) where the operating system is installed.
For this test, we performed an initial build of the image 'angstrom-lxde-image', including downloading and compiling all packages, as well as downloading the image from the cloud to the development computer.
I used a laptop computer from work for the test. It has an Intel® Core™ i7-3740QM CPU @ 2.70GHz with 8 virtual cores (the processor has 4 cores), 8 GB RAM, and an SSD for the builds, running Ubuntu 16.04 LTS natively. Starting from scratch, the initial build took four hours in all.
Next, two VMs were provisioned in Azure, as follows: one with the simplest configuration available (Standard DS1 v2), with 1 Intel® Xeon® core, CPU E5-2673 v3 @ 2.40GHz, 3.5 GB RAM and SSD disk; the other one with the most powerful configuration available to my account (limited to 10 cores), with 8 Intel® Xeon® cores, CPU E5-2673 v3 @ 2.40GHz, 28 GB RAM and SSD disk. The total time for the initial build was 12 hours for the simplest configuration, and 1 hour and 30 minutes for the most powerful one.
A coworker of mine also ran some builds using Amazon AWS machines (c4.2xlarge, c4.4xlarge and m4.2xlarge – for more info, please check here) and other hardware configurations (4 core i7-6700K CPU @ 4.00GHz, 32 GB DDR4, NVMe SSD and 8 core Xeon® CPU E5-2667 v4 @ 3.20GHz, 64GB DDR4, RAM disk), for comparison purposes. These results are also listed in the table below.
Note 1: There are more powerful VMs available to other account types.
Note 2: Compiled image download time from the cloud to the local computer is rather short (< 3 minutes) for an average download speed of 1 MB/s.
Note 3: While some cloud platforms, such as Azure, round the VM costs to the nearest minute, other platforms might round the costs to the nearest hour.
Note 4: Using a RAM disk is also likely to improve performance.
|Machine||Complete Build Time||Cost per Hour*||Total Cost|
Intel® Core™ i7-3740QM CPU @ 2.70GHz, 8GB RAM and SSD disk
1 core Intel® Xeon® CPU E5-2673 v3, 3,5GB RAM and SSD disk
8 cores Intel® Xeon® CPU E5-2673 v3, 28GB RAM and SSD disk
|AWS 8-vcore c
8 cores Intel® Xeon® CPU E5-2666 v3, 15GB RAM and EBS storage
|AWS 16-vcore c
8 cores Intel® Xeon® CPU E5-2666 v3, 30GB RAM and EBS storage
|AWS 8-vcore m
8 cores Intel® Xeon® CPU E5-2686 v4 or E5-2676 v3, 32GB RAM and SSD disk
|Local Server 1
Intel® Core™ i7-6700K CPU @ 4GHz, 32GB RAM DDR4 and NVMe SSD disk
|Local Server 2
Intel® Xeon® ES-2667 v4 CPU @ 3.20GHz, 64GB DDR4 RAM and RAM disk
* Hourly values presented in this table do not include international transaction fees. Costs reflect a VM deployed on the East US Azure server, and may vary depending on the region in which the VM is deployed. Other services, such as data storage, were used alongside the VM service; these may also be relevant to the total cost of the build. Prices may change without notice, per Microsoft's policy–please browse the Azure website or contact Microsoft for more information regarding pricing.
The simplest Azure virtual machine used costs US$0.073/hour, whereas the most powerful one (available to my account) costs US$0.585/hour. Therefore, the total first build cost is about US$0.878 for both machine configurations. The table of costs for all configurations can be found on this Azure page.
Tip #1: After you finish a build, shut down the VM to avoid spending credits unnecessarily. The idea is to pay per use, so remember not to leave a VM running over the weekend, for example.
Tip #2: You can configure automatic VM shutdown functions as a safeguard, in case you forget to shut down a VM manually.
This experiment has revealed both pros and cons for cloud-based builds. Given the build times from the "Is it fast?" section, a cloud-based build strategy can potentially speed up the build process significantly. However, you should also take into account download speeds: with a good internet connection, a low-cost local build server can be as fast or even faster than some cloud configurations. On the other hand, if you have a slow or unreliable connection, the cloud-based approach may be a better option.
Additionally, consider that the cost of using a VM over time might be higher than the cost of running a local build server. On the other hand, if your company does not have a dedicated local server for builds, using the cloud might save time for some employees. This is cost-beneficial, since VM hours are cheaper than employee working hours. This strategy also saves potential wear and tear on personal machines from running builds.
The cloud may also work well as an intermittent solution, to be used when multiple builds need to be performed simultaneously. This solution is easy and inexpensive to implement alongside a local server that is already handling one or more builds concurrently.
In the end, it is important to understand what kinds of cloud solutions are available, and to be aware that these solutions may or may not fit your needs. I hope you find the tips and case study presented in this article useful. Until next time!