blob: a6b009f21658a22a1e579931e0cdcfa3d248b4c8 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
|
==AutoScaling==
To get it to run --? look at how failures are implemented.
The moment your pool of usable hosts becomes too tight then you activate new ones.
1) I want to implement which allows me to put hosts inactive and activate it again.
2) During the workload the operator should be able to close and open the workloads.
i) then you build a policy as to when you activate new hosts and when you want to stop hosts.
ii) have many policies to do this
iii) add booting time --> cold start. A host becomes available but it takes still more time to make it workable because you need to boot everything.
iv) cold starts use zero power or very minimal --> idle power of a node that is turned off. Inactive have minimal power draw, but the downsides is that nodes may not be available immediately.
Inactive takes more time to boot e.g., 5 minutes.
The more you can predict how many nodes you will need to more you can save power without impacting the performance.
The bachelor focus -- the primary focus should be on the design of the digital twin.
Be careful that you do not make implementing the main topic of the bachelor thesis.
The main topic should still be digital twinning.
The digital twin is activated in 2 ways: either the operators prompt the digital twin when they run a workload or the digital twin sends a notification to the datacenter.
IBM Dublin := you have metrics being managed and policies can decide to do actions.
And these actions can be many different things.
Currently does not matter what it would -- this could be auto-scaling, scheduling, routing.
What is the difference between routing and scheduling?
Great success == we do not include AI inference in the bachelor thesis!
Policy decisions can be made using different heuristics --> AI inference.
Do not work too much on the policies.
The focus of the thesis is the digital twinning part.
Autoscaling vs. failures vs. scheduling.
MVP soon.
Why can't we port a digital twin from other domains?
Show that a digital twin can react.
Show the readers what would the perfect experiment look like in the perfect.
What a digital twin is --> give the answer in the background somehow.
What do you think a digital twin is? Make it clear.
Why is a specific datacenter digital twin different from what there already is.
Versen Thesis Awards they are promoted at ICTO today and tomorrow.
|
