Essay

Question 1

Difference between parallel and distributed computing:

Answer 1

Parallel
- Number of computers: 1
- Memory type: shared or distributed
- Communication: bus
- Goals: improve performance

Distributed
- Number of computers: multiple
- Memory type: distributed
- Communication: message passing
- Goals: improve scalability, fault tolerance, and resource sharing

Question 2

Difference between shared-memory and distributed-memory parallel computing:

Answer 2

Shared
- Programming: through threading
- Communication: Through shared pool of memory
- Pros: Easier to program
- Cons:
* Performance may suffer if memory is far from others
* Limited scalability

Distributed
- Programming: through processes
- Communication: message passing
- Pros: Tight control on message passing
- Cons: Harder to program

Question 3

Distributed transparency types and their descriptions:

Answer 3

Access: hides difference in data representation
Location: hides the location of an object
Relocation: hides the fact an object may move while in use
Migration: hides the fact an object may move
Replication: hides the copying of an object
Concurrency: hides that the object can be shared by independent users
Failure: hides the recovery of an object

Question 4

Dependability requirements and their descriptions:

Answer 4

Availability: Readiness for usage
Reliability: Continuity of use
Safety: Low probability of catastrophes
Maintainability: Easy to repair

Question 5

Difference between failure, error, and fault + examples:

Answer 5

Failure: A component not living up to its specifications. Crash
Error: Part of a component that leads to a failure. Bug
Fault: Cause of an error. Sloppy programmer

Question 6

Difference between fault prevention, tolerance, removal, and forecasting + examples:

Answer 6

Prevention: Avoid occurrence of fault. Not hiring sloppy programmers
Tolerance: Mask occurrence of fault. Make two programmers work on the same component
Removal: Reduce number of faults. Fire sloppy programmers
Forecasting: Estimating future faults. Estimate a recruiter’s chance to hire a sloppy programmer

Question 7

Grid computing layers and what they do:

Answer 7

Fabric: Provides interface to local resources
Connectivity: Has communication protocols
Resource: Manages a single resource
Collective: Handles access to multiple resources
Application: Contains actual grid applications in a single organization

Question 8

Exploratory decomposition vs Speculative decomposition

Answer 8

Exploratory decomposition:
* Output from a branch is unknown
* Parallel programs do more/less/same amount of work as serial programs

Speculative decomposition:
* Input at a branch is unknown
* Parallel programs do more/same amount of work as serial programs

Question 9

Static task generation strategy vs Dynamic task generation strategy

Answer 9

Static task generation strategy: Concurrent tasks and task graph known a-priori
Dynamic task generation strategy: Concurrent tasks and task graph are computed on the fly

Question 10

Static interaction vs Dynamic interaction

Answer 10

Static interaction:
* Timing and interacting tasks known a-priori
* Easier to code

Dynamic interaction:
* Timing and interacting tasks not known a-priori
* Harder to code

Question 11

Static mapping vs Dynamic mapping

Answer 11

Static mapping:
* Optimal mapping may be NP complete
* Tasks mapped to processes a-priori
* Requires task sizes

Dynamic mapping:
* Tasks generated at runtime
* Tasks mapped to processes at runtime
* Task sizes unknown

Question 12

Centralized dynamic mapping vs Distributed dynamic mapping

Answer 12

Centralized dynamic mapping:
* Processors designed as masters or slaves
* Slaves request tasks from master
* Higher chance of bottleneck

Distributed dynamic mapping:
* Processors made equal
* Processors send and receive work from others
* Lower chance of bottleneck

Question 13

Type of naming systems and their descriptions

Answer 13

Flat naming: Random string
Structured naming: Human-readable
Attribute-based naming: (attribute, value) pairs

Question 14

Name linking types and their descriptions

Answer 14

Hard link:
Name is resolved by following a path in a naming graph

Soft link:
A node contains the name of another so we have to:
1. Resolve the first node’s name
2. Get the name of the other node
3. Continue resolving with the other node’s name

Question 15

Difference between name-space levels

Answer 15

Global
- Directory node level: High
- Directory node administration: Jointly managed by different administrations

Administration
- Directory node level: Mid
- Directory node administration: Grouped in a way that can be assigned to separate administration

Managerial
- Directory node level: Low
- Directory node administration: Within a single administration

Question 16

LDAP essence:

Answer 16

Directory information base: Collection of all directories
Relative distinguished name for each record: Each record is uniquely named as a sequence of naming attributes
Directory information tree: Naming graph of the directory service

Question 17

Application layers and what they contain

Answer 17

Interface layer: Units for interfacing
Processing layer: Functions
Data layer: Data to be manipulated through components

Question 18

RESTful operations and their descriptions

Answer 18

PUT: Create resource
GET: Retrieve state of resource
DELETE: Delete resource
POST: Modify resource by transferring new state

Question 19

Coordination between temporally and referentially coupled

Answer 19

Temporally and Referentially coupled: Direct
Temporally coupled and Referentially decoupled: Event-based
Temporally decoupled and Referentially coupled: Mailbox
Temporally and referentially decoupled: Shared data space

Question 20

Multi-tiered centralized system architecture and their descriptions

Answer 20

Single-tiered: Dumb terminal/mainframe configuration
Two-tiered: Client/single server configuration
Three-sided: Each layer on separate machine

Question 21

Structured P2P vs Unstructured P2P + example:

Answer 21

Structured P2P
* Data represented as (key: value) pairs
* Keys used as indexes
* Example: Hypercube

Unstructured P2P
* Nodes store list of neighbors that looks like a random graph
* Searching for a node is done through flooding or random walk
* Example: Chord

Question 22

Cloud computing layers and their jobs

Answer 22

Hardware: Stores processors, power, and cooling systems
Infrastructure: Deploys virtualization techniques
Platform: Provides abstraction for storage
Application: Actual application like office suites

Question 23

Consensuses in appending a block to a blockchain and their descriptions

Answer 23

Centralized solution: An entity decides which validator can append a block
Distributed solution: A group of servers decide which validator can append a block
Decentralized solution: Participants elect a leader to append a block

Question 24

Sources of overhead and their descriptions

Answer 24

Interprocess interactions: Processors working on non-trivial problems need to talk to each other
Idling: Processors idle due to serial components
Excess computation: Computation not performed by serial version

Question 25

Execution time metrics and their descriptions

Answer 25

Serial runtime: Execution time of a sequential computer
Parallel runtime: Time from start of first processor to end of last processor