Day 3 - Embedded Systems and Technology Trends

📌 Key Takeaways

• An embedded system is a system built into a device for a specific function. It is found inside refrigerators, routers, and cars.
• Android is based on the Linux kernel. iOS is not — this distinction gets asked.
• Operating systems that use the Linux kernel — Android · Tizen · webOS · ChromeOS.
• The three cloud service models — IaaS (infrastructure) · PaaS (platform) · SaaS (software).
• The representative of big data distributed processing is Hadoop.

Embedded Systems

• Unlike a general-purpose PC, its job is fixed. So it is made small, containing only the needed functions.
• The reasons Linux is widely used in embedded systems are clear.

• An RTOS (Real-Time Operating System) is an operating system for systems that must respond within a fixed time. It is used where lateness is not allowed, such as a car's braking system.

What Embedded Linux Is Made Of

The week 2 boot sequence (bootloader → kernel → root file system) is there as is, only smaller.

• This is where why "open source" matters for embedded shows up. The kernel has to be trimmed and rebuilt yourself to fit in a small device, which is impossible without the source.
• An embedded device is too weak to build programs on itself. So we use cross-compilation: building executables for the target device (ARM) on a development PC (x86).

   Development PC (host, x86)               Embedded board (target, ARM)
   ┌─────────────────────┐    transfer      ┌──────────────────────┐
   │ Builds an ARM       │ ───────────►     │ Runs the ARM         │
   │ executable with a   │                  │ executable (does not │
   │ cross compiler      │                  │ build here)          │
   └─────────────────────┘                  └──────────────────────┘

Two Grades of Real-Time

• Real-time is not "fast" but "a guarantee of not being late". Even if the average is slow, it is real-time as long as deadlines are not violated.
• Linux is originally a general-purpose operating system, so it prioritizes throughput. Where strict real-time is needed, a dedicated RTOS such as VxWorks or QNX is used, or a real-time patch is applied to Linux. VxWorks and QNX are not based on the Linux kernel.

Operating Systems Based on the Linux Kernel

⚠️ iOS and macOS do not use the Linux kernel. They use Apple's Darwin (BSD family) kernel. Windows is of course not either. These names appear as wrong answers in questions like "which is not an operating system based on the Linux kernel".

• Android is the Linux kernel with its own framework on top. It shares the kernel but does not use the GNU toolset.

   ┌──────────────────────────────────────────────┐
   │ Apps        Phone · camera · downloaded apps │
   ├──────────────────────────────────────────────┤
   │ Application framework   Java APIs apps call  │
   ├──────────────────────────────────────────────┤
   │ Native libraries · Android Runtime (ART)     │  ← layer made by Google
   ├──────────────────────────────────────────────┤
   │ Linux kernel   drivers · memory · processes  │  ← layer shared with Linux
   └──────────────────────────────────────────────┘

• Android is the operating system of a company that Google acquired in 2005, and it was released together with the OHA (Open Handset Alliance), an alliance of manufacturers and carriers.
• The table above is the basis for saying in week 1 that "Android is not a Linux distribution". Only the kernel layer is the same; the rest is different.

Separate the names that get mixed into the choices into two boxes.

• What matters about the left box is that the makers are all different. Because the Linux kernel is released under the GPL, anyone can take it and use it as the foundation of their own operating system.
• Tizen and webOS survived in TVs and home appliances rather than smartphones. If a passage says "the operating system of a smart TV", it is one of these two.

Cloud Computing

   IaaS  ─── Rent only the hardware.  I install from the OS up
   PaaS  ─── Runtime environment provided.  Just deploy the app
   SaaS  ─── Use the finished product.      Nothing to install

• Tell them apart by the middle letter — I is Infrastructure, P is Platform, S is Software.
• The further down you go, the less there is for you to manage.

Spreading "who manages it" out by layer shows the boundaries of the three models.

• The boundary between IaaS and PaaS is the operating system. If "the user installs from the operating system up", it is IaaS; if "just deploy the app", it is PaaS.
• The boundary between PaaS and SaaS is the application. If you deploy a program you made, it is PaaS; if you just use a finished product made by someone else, it is SaaS.
• Unfamiliar names like DaaS (Desktop as a Service — a service for renting a desktop screen) sometimes get mixed into the choices. Just knowing it is not one of the three basic models lets you filter it out.
• The cloud is a service built on the virtualization we learned yesterday. One physical server is split into several virtual machines and rented to several customers. That is why "virtualization" appears in the bottom row of IaaS.

• OpenStack is a representative open-source platform for building private clouds.
• OpenStack was started in 2010 by NASA and Rackspace. As a service model it is a tool for building IaaS, and it is installed on Linux servers.
• Similar open-source projects include CloudStack and Eucalyptus. All three are not services for "renting the cloud" but software for "building the cloud yourself".

Big Data

• The characteristics of big data are explained by the 3 Vs — Volume · Velocity · Variety.
• Both Hadoop and Spark run on Linux.

Read Hadoop as two parts: storage (HDFS) and processing (MapReduce).

   HDFS — cuts big files into blocks and replicates them across several servers
 
     NameNode (remembers which block is where — holds only the listing)
        │
     ┌──┴─────────┬────────────┐
   DataNode 1   DataNode 2   DataNode 3      ← store the actual data
    Block A      Block A      Block A        same block, 3 copies by default

Following MapReduce through "word counting" makes the meaning of the name clear.

   Input       Server 1: linux unix linux     Server 2: unix linux
 
   Map        (linux,1) (unix,1) (linux,1)   (unix,1) (linux,1)     ← each splits and marks
   Gather      linux → 1,1,1                  unix → 1,1             ← same words together
   Reduce     linux 3                        unix 2                 ← summed up

• Map scatters and processes individually, and Reduce gathers and sums up. Because the computation happens right on the server that holds the data, there is no need to move big data to one place.
• Hadoop is an Apache Foundation open-source project and is written in Java. In that it ties together several cheap servers, it shares its roots with the cluster from day 1.
• Hadoop and the HPC cluster both "split and process", but the target differs. HPC is for big computation, Hadoop is for big data.

• The basic set is the 3 Vs (volume, velocity, variety). Plausible-sounding Vs like Virtual and Visual get mixed into the choices.

Where Linux Is Used

• The fact that Linux's share in the supercomputer field is overwhelming falls within the exam scope. It is because the source can be modified to fit the hardware.
• On TOP500, the world supercomputer ranking, all 500 systems have been Linux since November 2017.
• The reasons Linux is chosen in each field boil down to two — it costs nothing, and the source can be modified to fit. Embedded picks it to trim down, supercomputers to tune to the hardware, and the cloud to eliminate the license cost of countless instances.

Recent Technology Terms

• IoT is things such as appliances, sensors, and cars connecting to the internet and exchanging data with each other. What sits inside those things is the embedded system we learned about at the start of today.
• The characteristics of 5G are summarized as three — ultra-fast · ultra-low latency · massive connectivity. The body that makes the standard is 3GPP, exactly as covered in week 9.
• Edge computing is the opposite side of the cloud. Sending all data to a distant data center is slow, so it is processed first, close to the device.

• Kubernetes is an open-source project started by Google. If yesterday's Docker is the tool that starts one container, Kubernetes is the tool that places many containers across several servers and restarts them if they die.

Telling Them Apart from Passages

Commonly Missed Points

1. iOS is also based on the Linux kernel → No. It is Apple's Darwin. It only resembles Linux in being Unix-family.
2. Android is a Linux distribution → It shares only the kernel. It does not use GNU tools.
3. Tizen and webOS have their own kernels → Both are based on the Linux kernel.
4. An embedded system is a type of cluster → A completely different concept. It was a wrong choice in the day 1 question.
5. An RTOS is the operating system with the fastest processing speed → It is not about speed but a guaranteed deadline.
6. With PaaS you install from the operating system up → That is IaaS. PaaS gives you the runtime environment too.
7. SaaS is the model that rents out infrastructure → The order is reversed. I is infrastructure, S is software.
8. OpenStack is a big data technology → It is a cloud-building platform. Big data is Hadoop.
9. HDFS processes the data → HDFS is for storage; processing is MapReduce.
10. Kubernetes is a tool for creating containers → Docker does the creating; Kubernetes does the management (orchestration).

Expanding the Abbreviations

From tomorrow, we gather everything learned so far and work through comprehensive questions.

📖 Glossary

• Embedded system : A computer system built into a device for a specific function.
• RTOS : A real-time operating system that must respond within a fixed time.
• Android : A mobile operating system made by Google based on the Linux kernel.
• IaaS / PaaS / SaaS : Cloud service models that provide infrastructure / a platform / software.
• OpenStack : An open-source platform for building private clouds.
• Hadoop : A big data technology that splits large volumes of data across several servers to store and process.
• HDFS : Hadoop's distributed file system.
• MapReduce : A processing method in which data is split and processed on several servers (Map) and the results are gathered and merged (Reduce).
• Cross-compilation : Building, on a development PC, executables for a device with a different CPU architecture.
• Kubernetes : An orchestration tool that automatically deploys and manages containers.