Part1

(1) Distributed Transparency: For a client to communicate with a server using the server’s name, the communication needs to have:

• location transparency

(2) RPC Execution Semantics: Client-server communication middleware supports at-most-once RPC semantics because it simplifies:

• server implementation

(3) Inter Process Communication(IPC): In inter process communication(IPC), the communicating processes:

• should run on the same networked host.
• can run on different networked hosts.

(4) Inter Process Communication(IPC): Implementation of inter process communication requires:

• sockets
• protocol

Points:

IPC mechanism:

• Shared Memory
• Socket
• Semaphore
• Protocol

(5) Request-Reply Protocol: A remote procedure call(RPC) between a client and a server is typically implemented using a request-reply protocol because:

• The client does not know whether the server is running
• The server may not be running

(6) Remote Procedure Call(RPC): For a client process to call a local server(running in the same host) the same way as a remote server(running in a different networked host), we need:

• access transparency
• location transparency

(7) Local Procedure Call(LPC): In a local procedure call(LPC), the communicating(caller and called) processes:

• can run on the same networked host
• can run on the same standalone host.

(8) Data marshalling and unmarshalling are required in the implementation of

• both LPC and RPC

(9) Network Communication: The IP routing protocol to route a packet/message between a source and a destination:

• is a peer-to-peer distributed algorithm
• uses local information about the network

(10) Distributed Applications: An application is designed as a distributed application in order to achieve:

• transparency
• scalability

Points: Why Distributed Systems?

• Easily connect user to remote resources
• Share resources with remote users in a controlled way
  • Transparency: hide the fact that the resources are physically distributed over a network
  • Open System.
  • Scalable: Size; Geography and Administration.

Part2

(1) Middleware like RMI and Corba for distributed system achieve access transparency using

• Server interface definitions
• Stubs and skeletons

(2) A corba applicatiuon deployed on a LAN may not use

• implementation reposity
• corba naming service

A corba application in LAN use:

• object adapter
• interface definition

Points:

Corba Naming Service: is a core component in CORBA, it provides services of object naming and finding.

An implementation repository is a central location or system for storing and managing service implementation code

Object Adapter: play a role between the client and server, responsible for the initialization, delete, invoke and binding, encode, decode, transmission and respose.

(3) Corba application use Corba Data Representation CDR because they:

• can be implemented in different programming languages
• can execute on different hosts

Points: Common Data Representation: a coding scheme for marshalling and unmarshalling data of each IDL data type.

• functioins:
  • how to encode, marshalling and transmission.
  • transmission between different hosts, operator systems, and languages.
  • cross-platform, cross-language, cross internet.

(4) Implementation repository in RMI, a RMI distributed application does not require an implementation repository because RMI applications:

• do not use multiple language implementation
• know the location of server implementations

(5) When a distributed application is implemented in a same/single programming language and deployed on a LAN using JAVA RMI and Corba, the Corba Implementation will run

• slower due to middleware overhead.

Points:

RMI provides a simple, direct, and consistent API, which is integrated in JVM and Java Language.

Corba provides a cross-language, cross platform dirstibuted communication mechanism, which supports multiple programming languages and operating systems.

(6) A Java RMI application cannot be deployed on a WAN because of

• RMI Registry
• RMI Middleware

Points:

RMI Registry(Remote Method Invocation Registry): it provides a simple, light object registration services, which used for manage and maintain the remote object reference and address. RMI Registry is responsible for registing, finding, binding, and unbinding the remote objects.

• simple flat table that cannot be used on a WAN, since it’s not hierarchical

RMI Middleware: it is designed for the LAN, not for the WAN. WAN requires message exchange between networks, so it is not supported

(7) Location Transparency in Distributed Systems: In distributed systems, location transparency is typically achieved by:

• by the programmer using central directory

Points:

Location transparency: The client can access the server, without knowing where it is. Its difficult to do this because we always needed the know where the server is. So, its acieved by the programmer to add a central directory to the main server for the client to access by name.

(8) Corba Applications achieve platform independence:

• IOR and IIOP
• object adapters

Points:

Platform Independence: can execute in any hardware/operating system

General Inter-ORB Protocol(GIOP): a specification which provides a general framework for protocols to be built on top of specific transport layers.

Internet Inter-ORB Protocol(IIOP): is a special case of GIOP, which is the GIOP applied to the TCP/IP transport layer. The IIOP specification includes:

• Transport Management Requirements
• Definition of Common Data Representation
• Message Formats

The Object Request Broker(ORB): mediates the interaction between client and server objects.

As in Java RMI, a corba distributed object is located using an object reference. Since CORBA is language-independent, a CORBA object reference is an abstract entity mapped to a language-specific object reference by an ORB, in a representation chosen by the developer of the ORB.

Interoperable Object Reference(IOR): An ORB compatible with the IOR protocol will allow an object reference to be registered with and retrieved from any IOR-compliant directory service.

• CORBA object references represented in this protocol are called: “Interoperable Object References(IORs)”.
• IOR is a string that contains encoding for the following information:
  • The type of the object
  • The host where the object can be found
  • The port number of the server for that object

(9) Java RMI applications achieve platform independence using:

• Java Virtual Machines
• RMI middleware

Points:

Java RMI run based on JVM, which provides a abstract calculation environment and hides operating system and hardwares for users.

(10) A corba application implementation in a single programming language may not require the:

• implementation repository
• naming service

Points:

CORBA Naming Service: CORBA specifies a generic directory service, which serves as a directory for CORBA objects, which is platform independent and programming language independent.

• Name Resolving: The naming service permits ORB-based clients to obtain references to objects they wish to use. Name asscociate with the object reference.
• The API for the Naming Service is specified in interfaces defined in IDL

Part3 Review

(1) In a distributed application based on web services, it is desirable to implement the client:

• synchronously as the server response time is high.

(2) In web service technology, clients and servers communicate:

• synchronously by sending at the request and reply message asynchronously.

Points:

Web Services are software components described via WSDL which are capable of being accessed via standard network protocols such as SOAP over HTTP.

• SOAP provides rules for encoding the request and its arguments.
• WSDL documents are used to drive object assembly, code generation, and development tools.

(3) Unlike CORBA, a distributed application implemented in web service technology does not use an implemnation repository because:

• servers may be implemented in any programming languages
• requests and reply are encoded in XML.

Points: An implementation repository is a central location or system for storing and managing service implementation code.

(4) Unlike CORBA, interface definitions are not written by the server developer in web services because:

• they are generated from server end point.
• they are included in the service description.

Points:

Server Endpoints:

• clients can communicate with server by this Server Endpoints.

WSDL(Web Services Description Language)

• WSDL is a XML format language, used for descripe web services interface, operation, message and protocle detail.

(5) In web services, programming language independence is achieved by using a client and server implemented in any programming language, only when they can communicate irrespective of the language they are implemented in:

• SOAP message
• XML encoding

Points:

XML works as the encoding format.

• SOAP message use XML as its default encoding format.
• SOAP message contains Envelope, Header and Body.
• Each elements in SOAP message are encoded as XML.

(6) Unlike CORBA, a client and client-side middleware need not be in the same programming language in web services because:

• the client and the middleware communicate using XML
• the client and the middleware communicate using SOAP

(7) In web service technology, platform-independence is achieved by(application can perform on any software or hardware):

• using a web service middleware.
• encoding the requests and replies in XML.

(8) In web service technology, a client needs to discover the service because:

• the available service changes frequently.

Points:

The web services working process:

• First the client discover the service
• client binds to the server(binding not always need)
  • Setting up TCP connection to the discovered address
• Build the SOAP request(Marshalling)
  • Fill in what service is needed, the arguments, send it to server side.
  • XML
• SOAP router routes the request to the appropriate server
• Server unpacks the request, handles it, computes results
• result sent back in the reverse direction: from the server to the SOAP router back to the client.

Reposity:

• a database listing servers
• Each is described using the UDDI language, which is defined over XML.
  • can be searched with XML queries.

UDDI is used to write down the information that became a “row” in the repository.

WSDL documents the interfaces and data types used by the service.

(9) Unlike CORBA, stub and skeleton codes are not used in web services because:

• their functionalities are included in WSDL description.
• SOAP is used in web services.

stub and skeleton generate request and reply messages, so SOAP more relevant than XML

(10) The description of a web service in WSDL contains the service:

• interface
• reference

Part4 Review

(1) In an actively replicated system, it crash failures are detected using the absence of a result like in the project, the server:

• cannot detect without a client request
• cannot tolerate a software failure

(2) When all the replicas in an actively replicated server system execute a set of requests in total order. The local copy of the data in every replica will be identical after:

• each client request is processed.

Points:

Data in all nodes are actively updated and motified. When a node is received a write operation: insert, update, delete, it will update the operation to its copy immediately.

(3) In order to ensure data consistency in a passively replicaed server systemm the backups should perform the data updates send from the primary in:

• FIFO order.

Points:

In Passively Replicated Server, data only need to be copied to other nodes in specific time slot, it should not be copied each write operation.

If a primary recieved a write operation, it will reserve the operation, in some specific time slot, it will update the data to its replications. Asynchronous.

(4) In distributed application implemented using active replication in general, the server replicas should execute a set of a client requests in：

• total order
• casual order

Points:

Total Order: every events have a global, linear execute order. In a total order system, all nodes agree the event order in the time line.

Causal Order: If event A happens before B and influences B, all nodes should agree A is before B.

(5) In replicated Server System, each replica maintains a local copy of application data in order to:

• execute client operation faster
• access data in parallel

Points:

Actively Replicated:

Benifits:

• Performance Enhancement
• High Availability

Shortage:

• network cost
• consistent cost

(6) In an actively server system if the FE does not repliably multicast a client request to the server replica:

• the client request may not executed by all replicas.
• a software failure may not be detected

(7) In replicated server systems, the FE typically invokes a server method by sending UDP message because the FE:

• invoke the method in multiple process
• minimize invocation overload

(8) In an actively replicated server system, replicas should send the result of a client operation to the front end:

• reliable unicast
• UDP message

Points:

About reliable unicats:

• ACK
• Overtime and retransmission
• Guarantee sequence
• Error Check and Fix

(9) In an actively replicated server system, the replicas should be implemented as:

• iterative server

Points:

Iterative Server: when process the task, it will not start a new thread, but process the tasks in sequence.

• No concurrent
• Single task processing
• Simple and predictable

(10) When a software failure happens in an application implemented using passive replication:

• the primary replica produce incorrect result(because of the passive copy property)
• the local data in all replicas will be incorrect

Points:

Software Failure: refers to the abnormal or incorrect behavior of the system caused by software errors or defects.

• may lead to the incorrect of the data
• influence the primary and all other replicas
• need to fix errors

Crash Failure: refers to the sudden stop or crash of an application or system component due to a hardware failure, operating system error, or other system-level error.

• unpredictable
• influence single replica
• trigger the recover mechanism

Part5 Review

(1) Concurrent snapshots can be taken in a distributed system using Chandy and Lamport’s algorithm becasue:

• the markers can differentiate the concurrent snapshots
• the algorithm is non-blocking

(2) The Chandy and Lamport’s distributed snapshot algorithm makes a cut consistent by:

• including all the messages in the network at that instance of time in the channel state of a process.

Points:

instance of time in the channel state of a process: in a specific time slot, We want to know what messages are being delivered or waiting to be received in a particular process’s message channel.

(3) In a distributed system, a consistent cut represents the system states:

• that can happen during some execution

Points:

A consistent cut: something that can happen when the system executes.

(4) In Chandy and Lamport’s distributed snapshot algorithm, a cut is:

• the set of time points (one per process) at which the processes receive the marker for the first time.

(5) In distributed system, a cut is

• a collection of time points, one per process during an execution
• the current time in the distributed system

(6) If there is no path between some pair of process, the Chandy and Lamport’s distributed snapshot algorithm will:

• produce incomplete state
• not terminate

(7) If the communication channels are not FIFO, the set of states collected by Chandy and Lamport’s snapshot algorithm will be inconsistent because:

• a marker sent by a process may overtake a message
• a marker sent by a process may arrive before a message already in the network.

Points:

• Overtaking is corrupting message and overwriting in arriving earlier. (Check could be opposite.

(8) Global States in a distributed snapshot contain the appropriate information about

All the processes and the messages among them relevant to the snapshot.

(9) Distributed System cannot reliably use physical time because:

• It is not possible to make the times at all the hosts always the same.

(10) in distributed system, the vector clock maintained by a process corresponds to its knowledge about:

• the relevant events in all the processes from which it has received messages