1
Q
What is a distributed system?
A
- A collection of nodes that appear to users as a single coherent system
- Nodes can be hardware devices or software processes
2
Q
What are 4 benefits of a distributed system?
A
- Resource sharing
- Performance improvement (parallel execution)
- Scalability
- Reliability and availability (redundancy improves fault tolerance)
3
Q
Fallacies:
How is “the network is reliable” a false assumption?
A
- The network can have hardware failures or software bugs
4
Q
Fallacies:
How is “the networkis secure” a false assumption?
A
- Networks are vunerable and security must be built into applications
5
Q
Fallacies:
How is “the network is homogeneous” a false assumption?
A
- Systems contain different OSs, hardware and protocols
- Require interoperability
6
Q
Fallacies:
How is “the network topology does not change” a false assumption?
A
- Devices frequently join and leave
- Use abstractions like DNS instead of IP address
7
Q
Fallacies:
How is “latency is zero” a false assumption?
A
- Communication delay is unavoidable
8
Q
Fallacies:
How is “bandwidth is infinite” a false assumption?
A
- There is limited transfer capacity and data compression
- Traffic management is required
9
Q
Fallacies:
How is “transport cost is zero” a false assumption?
A
- Data serialisation requires time and infrastructure
- Network usage costs money
10
Q
Fallacies:
How is “there is one administrator” a false assumption?
A
- Multiple organisations manage networks and coordination
- This means that compatability issues may arise
11
Q
How does communicaton take place in a distributed system?
A
- Message passing: nodes send messages rather than sharing memory
- Marshalling: transforms object memory into a transferable format
12
Q
What is the middleware in a distributed system?
A
- The middle software layer above the OS that provides support and security
- Provides a uniform interface to application
13
Q
What is transparency in a distributed system and name 7 types?
A
- Transparency hides complexity from users and applications
- Types: access, location, relocation, migration, replication, concurrency, failure
14
Q
Explaing the difference between data and process synchronisation
A
- Data: keeping multiple copies of data in coherence
- Process: multiple processes need to act together for fast and reliable communication
15
Q
What is Lamport’s logical clock?
A
- When an event occurs, the logical clock is incremented by 1
- Logical clock is send to reciever, and reciever is set to one step ahead, to preserve partial ordering
- This means we cannot infer the order of 2 events
16
Q
What are central lock servers?
A
- Marks a resource as being used, and grants a mutex lock
- If not available, the mutex request is put into a queue
However Central Lock Server is a single point of failure
17
Q
What is a two-phase commit algorithm and give an example?
A
- Ensures a sequence of events is either completed or returned to original state
- Bully algorithm chooses a coordinator node which requests a transaction
- Sends election where all participating nodes must vote whether to commit or abort
18
Q
Explain the two clock synchronisation algorithms
A
- Cristians algorithm: uses a Round Trip Time RTT between nodes, then halves it, and adds it to the time servers time
- Berkley algorithm: server is master, averages RTT of other nodes and sends out amount each slave must adjust its clock by
19
Q
What are the 4 conditions of deadlock?
A
- Mutual exclusion: one resource is non-shareable
- Hold and wait: process is holding a resource
- No pre-emption: nothing can force a process to relinquish a resource
- Circular wait: process is waiting for a resource being held
20
Q
What is a naming resolution?
A
- How a process can access a named entity using its name
- Supported by a naming system
21
Q
Explain the 3 naming systems
A
- Centralised naming: single point of contact that allocates unique names- single point of failure and not scalable
- Free-for-all naming: allows object to make up its own name- not unique
- Delegating responsibilty: smaller part of system allocates names based on rules- balances issues with uniqueness and scalability+failure
22
Q
A
23
Q
Give 3 delegated naming systems
A
- MAC addresses (48-bit): unique identifier to each device- does not say where device is on network
- IP addresses (32-bit): contains information about where a device is on a network
- DNS: Distributed system on top of internet which associates names and IP addresses- uses real-time allocation with top of hierarchy dealing with user requests
24
Q
What is a protocol and give 2?
A
- Protocols are sets of rules governing how objects interact
- HTTP and SMTP
25
Explain HTTP
- **Hypertext transport protocol**
- Allows clients to **request resources** from stateless web servers
- Stateless means once request is fulfilled the server forgets the client
- Allows system to treat each request as an independent transaction
26
Explain SMTP
- **Simple Mail Transport Protocol**
- Connection based protocol that **sends emails** between clients and servers
- Ensures emails are in exactly one place at any one time
27
How does the OS manage resource usage?
- **Assigns unique identity** to each process and grants access by **assigning address space**
- Ensures no other process tampers with locations
28
What is multi-tasking?
- Allows multiple processes to be underway by **controlling resources**
- Uses a scheduling policy, **not really concurrent**
29
What is multi-processing and how does it differ in forking and threading?
- Multi-processing allows 2 processes to be active concurrently
- **Forking** **copies** a process and gives them distinct addres spaces- **safe but expensive due to copying**
- **Threading** **shares** the address space, rather than copying, which allows the processes to interact- **less expensive and less safe**
30
Explain the difference between concurrent, parallel and distributed?
- **Concurrent**: one processor with mutliple threads
- **Parallel**: many processors which interconnect
- **Distributed**: many independent machines that exchange messages
31
What is a centralised architecture?
- Traditional client-server where server implements software
- **Simple** only has **clients and server** with reply-request behaviour
- **Multi-tiered** has **UI layer, processing layer and data layer** which are seperated into client machine and server machine- all functionality handled by server
## Footnote
Called **vertical distribution**
32
What is a decentralised architecture?
- **Peer-to-peer architectures** where all nodes are equal
- Distributes **client-server functionality evenly**- better workload balance
## Footnote
Called horizontal distribution
33
What are overlay networks?
- Implementation of decentralised architecture
- **Symmetrical network** where each node communicates through other nodes
- **Structured**: nodes organised in a topology to efficiently look up data using hashing
- **Unstructured**: each neighbour requires an ad hoc list of neighbours, which changes continuously- requires searching
34
Explain 2 searching methods in overlay networks
- **Flooding**: nodes continuously pass request to all neighbours until it is found- **expensive** so **time-to-live value** needed to give max forwards allowed
- **Random walks**: randomly chosen neighbour passes along request- **less traffic**, but more time, so can impose n walks simultaneously
35
What is a hybrid achitecture and give 2 types?
- **Combines** centralised and decentralised architectures
- **Collaborative distributed systems**: centralised when a node joins, decentralised once it has joined
- **Edge-server systems**: deployed on internet, servers placed at edge of the network, serves all content
36
Give and explain 4 arhcitecture styles
- **Layered**: bottom layers provide service to top layers, with requests flowing from the top
- **Object-based**: objects correspond to components, and invoke methods to communicate (Remote Procedure Calls RPCs)
- **Resource-centered**: system is collection of resources that are managed by components and primary communication happens in data center
- **Event-based**: processes publish a notification and others subscribe to notifications
37
Explain the 2 characteristics of processed in event-based architecture style
- **Referentially decoupled**: one process does not explicitly know any other processes
- **Temporally coupled**: processes must be running at the same time to communicate
38
What is interprocess communication?
- **Communication mechanism** which allows processes on different machines to exchange info
- Enables data transfer and coordination
## Footnote
Forms basis of Remote Procedure Control and middleware
39
Explain Remote Procedure Call
- Hides message-passing by allowing processes to call procedures remotely
- **Client stub marshals** parameters into a message, **server stub unmarshals** and executes
## Footnote
Improves communication transparency, making it ideal for client-server architectures
40
What is parameter marshalling in Remote Procedure Call?
- Packing **parameters into message** before transmission
- Transforms data into a **machine-independant** and **network-independant** format
- Ensures data is interpreted consistently
- Passes references as objects by copying entrie data structures
41
Explain asynchronous Remote Procedure Call
- Server** immediately sends a reply/acknowledgement** after receiving request (before processing)
- Client completes execution without** waiting for full result**
- Enables **non-blocking communication** between processes
## Footnote
Useful in systems where reponsiveness is more important than immediate results
42
What is Message-Oriented Middleware?
- Middleware that enables communication through **message-queueing mechanisms**
- Supports **asynchronous** communication where **sender+reciever are not simultaneously active**
- Messages stored in queues+forwarded to destination
43
Year 2
flashcards
Decks in class (7)
# Cards
