exam_2021-retake

Question 1

What is the AFK scale cube?

Answer 1

It is a model for segmenting services, defining micro-services and scaling products.

Question 2

What are the axis’ of AFK scale cube?

Answer 2

X-Axis: Horizontal duplication and cloning of services and data
Y-Axis: Functional Decomposition and segmentation
Z-Axis: Service and data partitioning along customer boundaries

Question 3

What is horizontal duplication?

Answer 3

One monolithic system/service > many systems, each a clone and load balanced

Question 4

What is y-Axis

Answer 4

Split by function, method, service or dissimilar things

Question 5

What is Z-Axis?

Answer 5

Split by similar things. For example, customers

Question 6

Q1.A Explain the difference between service partitioning and data partitioning inthe AKF scale cube, and give an example of each

Answer 6

Splitting services is on the Y-Axis according to AFK scale cube, data partitioning on the Z-Axis. You can split services by separating the webserver from the database. You can do data partitioning by splitting a database (sharding) based on customer ID’s for example.

Question 7

Q1.B At one point, Google search introduced extra layers between the web frontend and the servers holding the partitions of the index. Why did they have to introduce these layers?

Answer 7

They introduced caching servers. These have a hit rate of 30-60% and are capable of handling a whole lot of traffic

Question 8

What is a race condition?

Answer 8

When the code is trying to do two or more things at once, and the result changes depending on the order they occur in

Question 9

What is one simple solution against race conditions?

Answer 9

To place all the requests in a queue and refuse to answer any requests until the previous one is completed

Question 10

What is the problem with placing all requests in a queue and refusing any further requests to prevent race conditions?

Answer 10

It doesn’t scale. That’s how old computers work, single-threaded.

Question 11

When is using a queue and refusing further requests a good solution

Answer 11

If you absolutely must count everything accurately, in real time. For example, a large festival with lots of requests where it can’t allow collisions.

Question 12

What is alternative to using this queueing system that does scale well?

Answer 12

Eventual consistency

Question 13

What is eventual consistency?

Answer 13

Each server holds it’s own count. It will update the central system when there’s time to do so. (it can also be seconds apart, doesn’t have to be hours).

Question 14

When is eventual consistency a bad design choice?

Answer 14

When a change needs to be made immediately. For example, the privacy settings of a youtube video (private or public)

Question 15

Why are youtube views not accurate?

Answer 15

This is because of caching. Caching holds the data and serves it to the customer quickly. A site like youtube has many, many caching servers and each time you can be routed to a different caching server.
Eventually, eventual consistency will take place and everything will be sorted out at some point

Question 16

Q1.C Google Search replaced batch processing to create the index to a more incremental method of keeping the index up-to-date. Why did they want to make this replacement?

Answer 16

They wanted to make this replacement because the batch process via MapReduce resulted in documents not showing up in search results for 2-3 days. They needed a lower “time from crawl-to-search-hit”. Solution was:

• New data storage system: Colossus / BigTable
• Event-driven, incremental processing: Caffeine / Percolator

Question 17

What is batch processing?

Answer 17

It gives you the ability to execute multiple operations in one request, rather than having to submit each operation individually

Question 18

Q1.D What problem is avoided by Google by time-stamping the contents of a BigTable cell?

Answer 18

Because of versioning by timestamps there are no write-write conflicts on a cell. As we will see: when replicated, eventual
consistency is used.

Question 19

Q1.E Briefly explain why virtualisation is considered a scaling technique?

Answer 19

It lets you run multiple operating systems and applications on a single server, consolidate hardware to get higher productivity from fewer servers and simplify the management, maintenance and the deployment of new applications

Question 20

Q2.A The parameters of a remote procedure call can be of 3 types: in, out, or inout. What are these?

Answer 20

in: object is transffered from client to service, only used for inputs
out: object is transferred from client to service only used for outputs
inout: object is transferred from client to service used for both inputs and outputs

Question 21

Q2 TODO

Answer 21

TODO

Question 22

Q3.A You are the administrator of an application that consists of web serverrs on 3 machines and a database on 1 (seperate) machine. The vendor of the application wants to change the application components (web server, database) from running natively to running inside a container. Discuss how this change would affect the performance of your application

Answer 22

There are some cases in which virtualisation offers performance benefits but these are quite rare, normally. Typically, the overhead of a hypervisor is around 1-5% of CPU and memory overhead.
By using some of VMware’s memory management techniques you can eliminate the memory overhead IF, and only IF, you are using multiple VM’s onto hosts. Using 1 VM will always be slower.
HOWEVER, while you lose some slight performance, virtualisation is about management. With virtualisation you can easily scale to 20 or 40 VM’s on each host

Question 23

Q3.B TODO

Answer 23

TODO

Question 24

Q3.C Classify firecracker into the virtualisation taxonomy discussed in class

Answer 24

These are system VMs capable of running a the same ISA (Linux)

Question 25

What is firecracker

Answer 25

Open source virtualisation technology that enables you to deploy workloads in lightweight virtual machines called microVMs which provide enhanced security and workload isolation over traditional VMs

Question 26

What is the virtualisation taxonomy?

Answer 26

You have 
Process VMs:
- Same ISA
- Different ISA
System VMS:
- Same ISA
- Different ISA

Question 27

Q3.D Give one similarity and one difference between firecracker and unikernels

Answer 27

The similarity is that they both offer significant performance improvements by excluding unnecessary devices and guest functionality.
The difference is that firecracker runs in user-space and unikernels in kernel-space

Question 28

What are unikernels?

Answer 28

They optimise a VM for one application. You strip unused parts of OS and libraries

Question 29

Where does firecracker run in?

Answer 29

It runs in user space

Question 30

What is gVisor?

Answer 30

Provides a virtualised environment in order to sandbox containers.

Question 31

gVisor advocates itself as third secure environment for running containerised applications, next to machine-level virtualisation and rule-based execution (such as SELinux). What disadvantages of the two others does it try to solve?

Answer 31

gVisor intercepts application system calls and acts as the guest kernel, without the need for translation through virtualised hardware.

Question 32

What are the different layers of gVisor?

Answer 32

1. Machine-level virtualisation (such as KVM, XEN), exposes virtualised hardware to a guest kernel via a virtual machine monitor
2. Rule-based execution, such as SELinux.
3. Intercepts application system calls and acts as the guest kernel, without the need for translation through virtualised hardware.

Question 33

4.A If we did not have dedicated configuration management tools like Ansible and Puppet, could an organisation still use Infrastructure as Code? Briefly argue why or why not?

Answer 33

Yes, IaC is the process of managing and provisioning computer data centers through machine-readable definition files. You don’t need a tool for this, it’s more about the methodology behind it which can help reduce cost speed and risks

Question 34

What are the advantages of IaC?

Answer 34

Cost
Speed
Risk

Question 35

What is cost advantage IaC?

Answer 35

By removing the manual component, people are able to refocus their efforts towards other enterprise tasks

Question 36

What is speed IaC?

Answer 36

Infrastructure automation enables speed through faster execution when configuring your infrastructure and aims at providing visibility to help other teams across the enterprise work quickly and more efficiently

Question 37

What is risk IaC?

Answer 37

Automation removes the risk associated with human error, like manual misconfiguration; removing this can decrease downtime and increase reliability

Question 38

Q4.B Some daemons reload their configuration files when they sense a change, causing downtime for that daemon. How do configuration management tools avoid this side effect?

Answer 38

I believe tools such as ansible have configurable parameters to specify whether daemons should be related. The default option is not to do so

Question 39

Q4.C Name 3 important pieces of information that IaC should record about a change

Answer 39

The change itself, the timestamp and the initial version

Question 40

How can you improve communication?

Answer 40

Chatrooms (force people to be online)

Virtual and physical standups

Question 41

Q4.D Explain why duvel beer is an important tool in DevOps

Answer 41

You want smaller changes to be happening throughout the week and not one massive change on a friday afternoon when everyone will go for drinks after?

Question 42

What is idempotence?

Answer 42

No changes when the same state is applied again

Question 43

Q5.A In your new job you are asked to migrate an “All Eggs in One Basket” server to setup with different virtual machines for each service. Briefly describe how you would perform such a migration?

Answer 43

Before migrating, I would ensure that I first create a backup. Then I would create a clone, so that you have the system still running, and a backup in case it goes down. Inside the clone I would setup the virtual machines which should be a machine each for each service to improve performance and it also gives management + security benefits. This should be done following IaC so that it’s reproduce-able. After the set up is fully working and tested, I would create a new VM to test if the whole configuration works. Then, the IaC can be used to configure a production instance (first shadowing the initial system). If all is well, it can replace the initial system. Again, a backup should be made of the new system.

Question 44

Q5.B The book advises against using desktop hardware to run services. Yet google and other big services have been built around such hardware. Briefly describe how one can build such services from desktop components.

Answer 44

1KW = 3412 BTUs
So, 3.52 x 3412 = 12010 of cooling
12010 / 1725 = 6.96
So you can cool about 7 servers (if, and only if, they are not all generating the full heat for a longer period of time, which is unlikely)

Question 45

How much BTUs is 1KW

Answer 45

1KW = 3412 BTUs

Question 46

Why do you, as an IT department providing a service, have to be very careful what goes into a SLA between you and a business unit in your company

Answer 46

The SLA is a commitment between a service provider and a client. It defines the level of service being sold in plain language terms with definitions about mean time between failures, mean time to repair or mean time to recover. You also identify which party is responsible for reporting faults or paying fees.