How to connect cloud to the cloudlet

[zetelight]

HI there, I'm Adam. After reading several papers from CMU, especially from Kiryong Ha, I am a little confused about how to connect cloud to the cloudlet in practice.

Definitions
At first, please let me double-check some definitions I understand in cloudlet model

"cloud"-"cloudlet"-"device".

Please correct me if I was wrong.

• Cloud: the cloud platform which provides cloud instances or clusters, such as OpenStack, Amazon AWS, Microsoft Azure.
• cloudlet: an additional tier in the model. In the repository, it is an extension of OpenStack.

Reasons why confused
When I installed everything from elijah-openstack, I thought "cloud" here should be OpenStack and "cloudlet" should be the extension. The elijah-openstack is the combination of "cloud" and "cloudlet" parts.. However after reading more papers, I felt like that the elijah-openstack itself is the solely "cloudlet"

Go back to the main question

• how to connect cloudtet to the cloudlet in practice? Spefically, I have "cloudlet" on my machine,
  how to reproduce the "cloud-cloudlet-device" model?

I have so many words to say but I don't want to dump tons of message to bother you. My current idea is following:

1. Use one EC2 instance(e.g. Ubuntu server) as my "cloud"
2. Use my local machine as cloudlet
3. Build some applications on another machine so that it can send requests to cloudlet or cloud to do some computing. I also can observe the performance in this way.

Am I right? Is the right way to use? Is there more paper for me to help me understand the how to reproduce them?

Some papers I've read

1. OpenStack++ for Cloudlet Deployment
2. The Impact of Mobile Multimedia Applications on Data Center Consolidation
3. The Case for VM-Based Cloudlets in Mobile Computing

Appreciate your patience and time so much!
Best wishes,

Adam

[teiszler]

Hi Adam (@zetelight),

You can think of the cloud and the cloudlet as being identical except for the proximity to the user. You can imagine that you have the same backend application running in Amazon EC2 in Oregon or Virginia and also one hop away on your local cloudlet. The frontend is running on some mobile device (could be a phone, a tablet, or some sort of wearable device like a Gear VR). The application is identical in both places, however when the user connects to the EC2 backend, the latency is much higher because the packets have to travel through the internet backbone all the way to the centralized compute center where you have launched your instance. If the user connects to the cloudlet, the experience should be better because of the reduced latency that is incurred by placing the compute at the edge, closer to the mobile device (in your environment this is likely a laptop or desktop machine that is on the same WiFi network as your mobile device, but in theory the cloudlets could be dispersed in many locations such as at the base of cellular towers, in public areas like coffee shops, and certainly onbaord inside of vehicles).

You could also envision applications where the three tier model you discuss is applicable. For instance, you might have an application that takes real-time video from a mobile device and ships it to a cloudlet that is nearby for processing against some object detection algorithms. These algorithms may require a GPU in the cloudlet and have a low latency bound because they return information to the mobile device. But they may also send the same information or various metadata about the objects detected to a service running in the cloud which is used to do aggregation or analytics.

So I think to answer your question about the cloudl-cloudlet-device model, it really depends on the kind of application you are thinking about (or perhaps already have) developing. There are certainly classes of applications, such as wearable cognitive assistants(http://elijah.cs.cmu.edu/DOCS/chen-sec2017.pdf), which require cloudlets to do low latency processing of video streams and provide tolerable feedback to a mobile user. There are other classes of applications that might have one component that resides on the cloudlet and another that resides in the cloud such as live synthesis of vehicle sourced data (http://elijah.cs.cmu.edu/DOCS/hu-livemap2017.pdf)

Hope this helps you get a better understanding.
Regards,
Tom

[zetelight]

Thank you, Tom. Appreciate your detailed explanation!
Now, I realized that why I got the wrong concepts. Previously, I always believe that cloudlet will only undertake some small workload from devices and send large workload to the cloud which has the better computing ability, so that's why I ask "how to connect cloudlet to the cloudlet".

Your words,"You can think of the cloud and the cloudlet as being identical except for the proximity to the use" really helps me understand it.
Personally, I think, compared with the classic cloud, most two unique values of cloudlet are

1. more lightweight, so it can use fewer resources to deploy VM/do VM handoff/do VM synthesis
2. closer to users.

Again, thank you Tom!
Best wishes,

Adam