diff options
Diffstat (limited to 'script/main.tex')
| -rw-r--r-- | script/main.tex | 78 |
1 files changed, 78 insertions, 0 deletions
diff --git a/script/main.tex b/script/main.tex new file mode 100644 index 0000000..a524e96 --- /dev/null +++ b/script/main.tex @@ -0,0 +1,78 @@ +\documentclass[12pt, a4paper]{article} +\usepackage{palatino, enumitem, parskip, xspace} +\usepackage[dvipsnames]{xcolor} +\newcommand{\eg}{\emph{e.g.,}\xspace} +\newcommand{\todo}[1]{\textcolor{Blue}{\textbf{TODO(#1)}}} +\newcommand{\etal}{\emph{et~al.}\xspace} +\begin{document} +\begin{center} + \Large My BSc Defence Script +\end{center} +\begin{enumerate}[label=\textbf{Slide \arabic*.}] + \item \textbf{Introduction}\\ + Good morning everyone, my name is Mateusz and today I will present to you my project \emph{Sunfish: Enabling Predictive Analytics For Datacenters Through Digital Twinning}. + + At a top level, my project is about trying to ease datacenter management by trying to pave the way to predicting unexpected events. + + \item \textbf{Societal Impact}\\ + As you know and as you will likely see in the upcoming presentations today, datacenters are important. + But, I would like to shortly mention this myself. + + + + A single GPU is already very complex. + Within a Google Datacenter, there are hundreds of server racks, with tens of such GPUs. + This begs the question: How are we going to manage this large of a datacenter, that has so many \emph{layers of complexity}? + + We cannot let these systems go down or experience big failures, because \eg in Netherlands over 3 million professionals depend daily on the cloud. + \todo{Read the slide box.} + As such, we must do something to manage datacenters well. + + + \item \textbf{Problem Statement}\\ + Digital Twinning pairs complex objects (like datacenters) via a two-way connection with their virtual representation. + \todo{Give example about the airplane from aviation.} + \emph{It a method to manage complex systems.} + + However, in digital twinning, specifically datacenter digital twinning a lot of elements are still shifting about and there are a lot of ways to create the virtual models and there seems to not be a fully functioning DCDT out there (\emph{that meets the official NASM definition}). + + DCDT's lack mandatory features one of which is predictive analytics. + Predictive Analytics is a type of ODA that draws insights into the future based on current data, \eg telling when a host failure might happen before it does (\emph{and yet it is NOT present in existing DCDTs}). + + \item \textbf{Research Questions}\\ + We wish to enable the development of predictive analysis components for DCDT's by designing a predictive DCDT. + We ask the following research questions. \todo{Read from slide boxes.} + + \item \textbf{Literature Survey}\\ + \item \textbf{System Model}\\ + + \item \textbf{Reference Architecture and Prototype}\\ + A reference architecture. + The design is more specific than the reference architecture. + + \item \textbf{Novel Evaluation Method}\\ + In order to evaluate a prototype, we propose a novel approach. + Many researchers do not have a real facility to experiment with. + We propose to use a second simulator to act as the real datacenter. + + \todo{Say in order to not cram content into the presentation, we omit the technical setup, and include it in extra slides.} + + + \item \textbf{Experiment 1: Red and Yellow Alarms}\\ + For Experiment 1 we copy the idea of Milojicic \etal for different ways a DCDT can notify the datacenter. + + Imagine a scenario: a datacenter will soon run a workload. + We want to detect and differentiate between failures that are big and unexpected and failures we anticipated would occur. + + To achieve this: the DCDT runs the workload using the simulator. + We cannot know what kind of failures we can expect, so we use a statistical distribution to approximate what might occur in practice. + In result, we get a picture of what kind of problems we might expect. + + We now use the real-time feedback loop to notify the DC operators that what is happening in reality is different from simulation. + If we get within 80\% of the predicted threshold for number of failures we send a yellow alarm. + If we get within 90\% we send a red alarm. + \item \textbf{Experiment 2: Conceptual Experiment}\\ + \item \textbf{Key Takeaways}\\ + \todo{Read from the slide.} +\end{enumerate} +\end{document} |