Client Server Architecture

Key concepts:

- client server architecture

- Functional requirements in the 2-tier structures
- Functional distribution in the 2-tier structures
- Implementation of Business Logic at the Server
- Requirements of an Open OLTP System
- Benefits and Limitations of 2-tier Client/Server Architecture
- The 3-tier Structures
- Middleware
- Component Software Model
- Database Middleware
- Transaction Processing and Integration Middleware
- Transaction Processing Monitors
- Performance Issues in TM
- Two-Phase Commit Protocol in TM
- Message Sensitive Routing
- Lifekeeper Clusters
- Repositories in Client-Server Environment
- 4GL Application Development Environment
- Communication Models
- The main standards for open transaction processing
Client - Server Architecture [Salem 1992]

The data processing is split into distinct parts. A part is

either requester (client) or provider (server). The client
sends during the data processing one or more requests
to the servers to perform specified tasks. The server part
provide services for the clients.

Client Server
request

service

This basic structure is called 2-tier structure

The main basic communication modes:

- RPC, remote procedure call
- Message-based

- The client and server parts may reside on the same node or on
different nodes
- A part can play the roles of a server of a service and a client of an
another service at the same time
- A client can be connected to several servers
Functional requirements in the 2-tier structures

The applications may be divided into the following logical functional

components:

- Presentation Services
user interface, dialog control
- Presentation Logic
user interaction, simple validation
- Business Logic
control flow
- Distribution Services
communication management
- Database Logic
integrity, data manipulations
- Database Services
security, transaction management
- File Services
file sharing

DESKTOP HOST
Data Logic
Client Server Distribution
Business Logic
File Services

DESKTOP Presentation
Client Business Logic
Data Logic
Functional distribution in the 2-tier structures

Fat clients: Most of the functional modules of the application are

performed on the clients
File server

Lite clients: Only few functional modules of the application are

performed on the clients

Remote Data Client-Server Architecture

Client Server
Presentation Data services
Business logic File services
Data logic

Remote Presentation Client-Server Architecture

Client Server
Presentation Business logic
Data logic
Data & file services

Split Logic Data Client-Server Architecture

Client Server
Presentation Business logic
Business logic Data logic
Data logic Data & file services
Implementation of Business Logic at the Server

Active elements in the Databases

- triggers
- stored procedures

Trigger:
stored DBMS procedures that are executed when there is change in
database.
- triggering event
- response
It is stored in the DBMS
Centralized management and access control

Oracle SQL:
CREATE TRIGGER name AFTER | BEFORE INSERT |
UPDATE | DELETE ON table FOR EACH ROW
BEGIN
PL/SQL block
END

Stored procedures:
collection of SQL statements and procedural language statements
that control the flow of the procedure
It is stored in the DBMS
Centralized management and access control

It provides faster execution

Mostly pre-compiled
Requirements of an Open OLTP System

Transactions
Logical unit of data processing operations
- Atomicity
- Consistency
- Isolation
- Durability

OLTP
On-Line Transaction Processing
Characteristics:
- Short time transactions
- Several concurrent transactions
- Read-write transactions
- Database stores the actual state of the problem domain

Requirements of open OLTP systems

(based on X/Open DTP model)

- Vendor independence
- Application portability
- Distribution transparency
- Modularity
- Scalability
- Reliability
- Reconfigurability
- Monitoring
Benefits and Limitations of 2-tier Client/Server Architecture

Benefits
- Interoperability
- Portability
- Integration
- Transparency
- Security

Limits
- The client and server are tightly coupled, the client should use the
interface protocol of the server.
- Network traffic is handled less efficiently, because it clients have
a direct connection to the server.
- The network traffic increases significantly when high volumes of
data and messages are shipped on the network.
- Asynchronous activity. - when a request is sent and an immediate
reply is not required - is not supported.
- The connections are limited to a single resource per transaction.
- The application administration is difficult.
- One cannot tune the response times and there is no capability to
perform load balancing
- The resources are scaled on the number of connected users not on
the throughput
- It consumes many operating system processes and network
connections in order to support a given set of users
The 3-tier Structures

Distribution of the different functional modules of the application

on three different sites

network
application
server

client

DB
database server

Usual distribution:
1. tier: Presentation logic, lite client
2. tier: Business logic, application server
3. tier: Database logic, database server

- less software on the client

- increased security
- more power, efficiency
- higher scalability
- lower support cost
- complex structure
- problem of heterogeneous data sources

The 3-tier structure can be extended to n-tier structure, containing

several special application servers
Internet-based 3-tier architecture

Open, standardized interface

network
WEB
server
WEB
browser

DB
database server

The WEB servers can access the databases through CGI interface

Interfaces among the components:

1./4. WEB browser WEB server
HTML API
<FORM . ACTION= URL>
.
</FROM>

2. WEB server DBMS

CLI (usually SQL)
SQLEXEC(SQL-statement)

2. DBMS WEB server

result written to standard output
HTP.PRINT(HTML-statement)
Middleware

In order to couple the server and client parts from heterogeneous

environments in an efficient way, the basic client server
architecture is extended by a new component.

Middleware
An integrating resource between the clients and servers. It performs
the following main activities:
- translation between the different protocols
- optimization of the load-balancing
- security control
- management of the connections
Server
Client
Middleware

..

Client Server

The middleware may contain several components.

The components may reside on the server node, on the client node
or on a new middleware node.

The different types of middleware

- database
- network
- application cooperation
Component Software Model

The clients invoke managed server components, the named

services. These services implement the core business routines.
This structure is based on the three-tiers model:
client: it performs GUI and invokes server components
middle tier: it is composed of server components to be invoked
by the clients
- control module with control data
- named services
resource manager: database management

Named services are managed dynamically, as it offers a better

optimization, starting and stopping the services depending on the
system load. These services may invoke other services.

Client Middleware DB
Control module Server

Service Service
A X

software pipelining:
A service processes only a part of the request and forwards it to an
another service for further processing. If the first service will be
free it can start to process the next request waiting for it. Thus the
requests are processed in a pipeline. If the request does not need
further processing steps, the actual service will return the result to
the client. A request will touch several services.
Control Module and Control Data

Function components of the control module:

- naming services
- message routing services
- load balancing services
- configuration management services
- transaction management services
- security services
Main system configuration data:
- system resources, security level, load balancing level,
- participating server definitions
- accessible services
- location information of the services, servers
- runtime repository of application statistics
Data dependent routing:
The service request is mapped to a specific server group based on a
value contained in a designated field. It routes the request to a
specific service/resource sets. The routing information are stored in
BB (Bulletin Board)

Security
It provides application service authentication , authorization and
access control through an architected secury interface. The
interface usually abstracts the Kerberos security model and allows
the security systems to be integrated with the application. Access
control lists may be used to protect services, queues, or events
from unauthorized access.
Database Middleware

The two main services of the database middleware are the data
management and the distribution services.

Data Management Service

It provides the transparent read and write operations to the server
database systems. When it receives a request for data, this service
translates the request to the appropriate database language and
performs a query on the target database.
This service makes also the database replication transparent to the
applications. It takes the responsibility of keeping all replicated
databases synchronized.

Distribution Service
This service provides distribution or database location
transparency for the applications. It unburden programmers from
needing to know where applications and data are stored. The
programmer has only to give the object name.
The real position of the object is determined by the directory
service. The directory service keeps track of real addresses of the
alias names. This enables scaling of applications by relocating
logic or databases onto more powerful machines.
Server
Rel. A
Client
Database T(Q(A))
Q(A),Q(B) Middleware
T'(Q(B)) Server
Rel. B
Transaction Processing and Integration Middleware

Transaction processing
The transaction processing (TP) is the automated, interactive
processing of transactions to update the a shared database. A
typical TP application features many end users interacting
concurrently with a system to process business transactions.

Transaction Processing and Integration Middleware

Integration software in distributed, heterogeneous environment. It
provides:
- Integration across heterogeneous platforms
- Reuse of existing applications and data
- Staging the activities involved in application and business
process re-engineering

trans:
Client
update B Oracle
update A Rel. A
Client
Transaction
Processing and
trans: Integration
Middleware
update A Sybase
update B Rel. B
Transaction Processing Monitors

A Transaction Processing Monitor (TM) manages and coordinates

the flow of transactions through the system. It guaranties the
ensuring of the ACID concept, the integrity of data accessed
across several sites or managed by different database products.

The TM coordinates distributed transactions to enable multi-site

updates against a heterogeneous database on networked computers.
It tracks transaction participants using global transactions and
supervises a two-phase commit protocol. It also coordinates the
recovery in the event of a site failure, network failure or resource
deadlocks. It uses algorithms designed to minimize the cost of
database accesses and network traffic.

Transaction at Client
Op.1 Op.2 Op.n
Global transaction
TM

Local transaction
Server A Server B
Op.2 Op.1 Op.3
Transactional RPC :
RPC's that carry transaction contexts along with their messages and
data, and that provide the exactly once invocation semantics
required for reliable distributed communications. It is extension
with transactional syntax.
Performance Issues in TM

Load-balancing algorithms can spread work over machines and

dynamically add more processes when hotspots of activities occur.
- static
- dynamic

To ensure maximum application throughput, the control module

automatically performs load balancing and scheduling. It delivers a
particular request to the server that can process the request most
quickly. It calculates the load on a given server by totaling the load
factors for the currently enqueued requests.
The assignment of the request is usually based on the actual load
distribution of the system.
Other, more sophisticated methods may be applied that also take
other aspects into consideration as
- network traffic
- task overlapping
Service A
TM
Server
Op j

Service B

TP monitorsfunnelling capabilities allow large number of clients

to share scarce system resources on nodes. This idea of sharing is
based on the statistical fact that not all clients want to perform
transactions exactly at the same time.
Two-Phase Commit Protocol in TM

In the case of distributed transactions, the global transaction

contains several local transactions.
The global transaction is successful only then, when all of the local
transactions are successful.
The Two-Phase Commit (2PC) protocol ensures this requirement.

One node among the participating transaction nodes should play

the role of a coordinator node. The coordinator node is responsible
for taking the final commit or abort decision.

2PC has two stages:

- a prepare phase
- a commit phase.
The purpose of the prepare phase is to allow the transaction
manager to collect a "yes" or "no" vote on the transaction from all
participating databases. If any "no" votes are received then the
entire transaction is rolled back. If all "yes" votes are received then
the transaction manager enters the second, commit, phase of the
protocol.
In the second phase a request to complete the transaction is issued
to each participating database. At the end of the second phase the
transaction is completed on all databases.

This ensures that in a distributed transaction that all servers will

either commit or rollback; never will one commit and the other
rollback. This is very important in the account debit example cited
earlier where one account must be credited and the other debited.
Recoverable Transaction Queuing, or RTQ accepts and processes
transactions using the store-and-forward concept: client
applications store transactions in a queue, and those transaction are
later pulled from the queue and processed. RTQ supports various
scheduling options for queued transactions

RTQ runtime management system architecture.

Client Database Server

Foreground application Background application

RTQ Server Queue RTQ Worker

RTQ is used to process transactions that do not require immediate

processing. A foreground application submits transactions to the
RTQ Server and a background application processes queued
transactions with the RTQ Worker. Queued transactions can be
processed at a scheduled time or in a trickle-feed fashion. For
example, to increase data protection an application can commit
database changes to a regular database and to RTQ. RTQ will
later forward these updates to an alternate (possibly off-site)
database.
Message Sensitive Routing

Caching:
When a client application reads a database record, the DBMS
executes an I/O to read that record from the shared database. When
the I/O completes, the DBMS retains a copy of the record in
memory. If any other application requests the record it can be
simply forwarded from memory and does not need to do an I/O.

A good performance can be achieved by a good "cache hit rate"

The distribution in DDBMS can cause a decrease of the cache hit

rate.

Pinging:
The same database record is updated at different sites, servers after
each other. Updated records that are released to another server
must first be removed or "flushed" from memory, as it is not
ensured that this value in memory remains the actual value in the
database due to the other server.
When pinging occurs frequently it can significantly reduce the
OPS cache hit rate and overall system performance.

The Message Sensitive Routing (MSR):

The requests for selected data always get routed to the same
DBMS server, thereby reducing pinging. MSR routes messages
based on a value (or set of values) within the request itself.
MSR improves the cache hit rate.
Lifekeeper Clusters

If there is a failure on server node in the DDBMS, in order to

resume processing it is necessary to first restore the server to
operational status and then restart the local DBMS and associated
applications. When restarted, DBMS performs recovery to restore
database integrity.

LifeKeeper:
is a software product that monitors two or more servers, detects
server failures, and automatically restarts failed applications on an
alternative server.
Because of its distributed service location transparency, the
Transaction Monitor can route requests accordingly if LifeKeeper
restarts failed applications on a different server.
TM
Server Server
A C

Lifekeeper connection Server

Service redirection B Server B fails

LifeKeeper service runs on each server and maintains a heartbeat

connection so it can detect and respond to server failures. The TM
replicates application logic within and across servers for scalability
and ensures continued access to services from either server in case
of component failures.
Repositories in Client-Server Environment

Repository:
a special database of information about engineered artifacts, that
holds the description of objects stored in normal databases

Repository manager

Database manager

Content of repositories:
- database objects definitions
- documents
- source code
- type definitions
- interface definitions
Menus
Forms

Repository
Reports
DB designer
Projects

Benefits of shared repository

- centralized access
- only one instance is required
- minimal redundancy
Repository products

- repository manager
- browser
- information model
- scripting language
- data import and export utilities

The repository manager supports the operations on the repository

including
- functions to manage change objects
- DB systems amenities

Some problems related to repositories

- managing changes
manage parallel development
- managing relationships
relationships among heterogeneous sources
- source and target type checking
integrity checking
- propagation of changes
operation on source propagates to target
- version management
dependency among the versions
- team development
parallel development, release management
- reusing
searching, conversion, re-definition
4GL Application Development Environment

The Transaction Managers may be integrated with the participating

DBMS's 4GL components.

Main 4GL components:

- Forms designer
- Menu designer
- Report designer

This integration has two levels:

- Interface module
- Generator module

TM Interface:
The TM Interface translates the TM APIs into DBMS native calls
accessible to the 4GL developer. The developed applications
benefit from the 4GL tools - provided features like GUI controls,
modal and modeless windows.

TM Generator:
This reads TM distributed services information and generates
DBMS 4GL objects. Once a service is selected, the default 4GL
modules are generated with objects corresponding to the TM
distributed service input and output parameters. The developer can
personalize the generated 4GL modules using the standard 4GL
components.
Conventional API of TEXUDO

The API provides interface for

- sending messages to a named service
- receiving results
- transaction demarcation
with four communication model:
- synchronous and asynchronous request/response
- conversation
- store and forward message queuing
- event based programming

TEXUDO : Distributed Transaction Monitor with high level API

Two basic types of API interface:

- library-based
- language-based

Library-based interface:
Library-based programming is supported via a set of 3GL (like C,
Cobol) procedures. The set of these procedures is called as
Application to Transaction Manager Interface (ATMI).ATMI is a
superset of the X/Open's XATMI interface.

Language-based interface:
The language-based programming is a remote procedure call
(RPC) facility called TxRPC. TxRPC defines a transactional
extension to the OSF's DCE RPC.
Communication Models

Library-based request/response mode

based on native functions
- tpcall(): calling service in synchronous case
- tpacall(): calling service in asynchronous case
- tpgetrply() : requesting the answer in in asynchronous case
- tpforward() : send the request to an another service

In synchronous case the caller resumes the execution until the

answer is received. In asynchronous case the caller does not wait
for the reply. It go on with the processing. The requested services
may be executed parallel. The caller can by means of the
tpgetrply() function synchronize the requested services.

{ tpinit(NULL);
data1 = tpalloc(BUFTYPE, SUBTYPE, len);
tpbegin();
tpcall('DEPOSIT', data1, 0, &data1, &len, 0);
tpcall('WITHDRAW', data1, 0, &data1, &len, 0);
tpcommit(0);
tpterm(); }

{ tpbegin();
handle1 = tpacall("withdraw', );
handle2 = tpacall('deposit'..); //parallel execution of services .
tpgetrply(&hand1,&r1 );
tpgetrply(&hand2,&r2 );
tpcommit(); tpterm(); }
Language based (IDL) request/response mode
It is based on TxRPC
RPC : remote procedure call, it uses function call formalism
The requested function code is located on a remote node and the
execution is performed on that remote node.
It can be only synchronous. The client suspends its execution until
the result is received.

{ tx_begin();
ret1 = withdraw(account1, amount1, &balance);
ret2 = deposit (account2, amount2, &balance);
if (ret1 == SUCCESS && ret2 = SUCCESS)
tx_commit; else tx_rollback(); }

Software pipelining
The requested service (function) may perform only a subpart of the
work requested by the client. In this case, it sends a request to an
another service t complete the task.

Client site calling a service: tpcall ('validate',data, )

Server site the implementation:
validate(R * reques){
data = request->data;
if (check(data.id) == SUCCESS) {
// forward to an another service
tpforward('CUST_STATUS',data, );
} else { raise_failure(..); tpreturn(failure,..);}}
Conversation-based mode

based on the following functions

- tpconnect() : client creates a connection to a service
- tpsend() : send a message to the partner
- tprecv() : reading the message from the partner
The server is called immediate, it uses a tprecv() to read the next
message from the incoming request queue.
Client site:
rec>bc = 100; /* to request 100 records */
cd = tpconnect('GETRECORDS', );
while(1) {
tprecv(cd,&recods, );
num_cnt = process_records(records,..);
if num_cnt < 100 break;
tpsend(cd, .); /* requesting the next set of records */
}
Service code:
GETRECORDS( *msg){
data = msg->data;
while (act_cnt < data->request_cnt) {
get_next_batch_of_records( );
tpsend(msg_cd,data, );
// wait for client that asks for more records
tprecv(msg->cd,&data, );
}
get_last_batch_of_records();
tpreturn (TPSUCCESS,..);}
Event-based mode

A server informs the clients about the happening of some events.

The clients are ready to activate a service-module as response to
the event.

The event-based communication can be:

- unsolicited : the clients register a service to each of their
events. If an event occurs, the registered service should be
called. Functions:
- tpnotify() : send message to only one client
- tpbroadcast():send message to one or more clients
- brokered : There is no direct connection between the even
generator and event consumer. There is three types of roles:
- subscriber: it subscribes at the broker for some events
- event generator: generates the events
- broker: it decides upon the subscription which clients
should be informed about the event
The main functions:
- tppost(): inform the broker about an event
- tpsubscribe(): subscribe an event
- tpunsubscribe() : unsubscribe an event
Event-based mode

Example

client site:
tpsetunsol (atm_handle);
// it tells the server to invoke the atm_handle function when an
unsolicited message arrives

server side:
the VALIDATE service
validate(R * reques){
data = request->data;
if (check(data.id) == SUCCESS) {
tpforward('CUST_STATUS',data, );
} else {
// notify the consol about the failure
tpnotify(consol, data,FLAG,..);
raise_failure(..);
tpreturn(failure,..);
}
Store and forward message queuing mode
based on the following functions
- tpenqueue() : place a message into the queue
- tpdequeue() : take a message out of the queue

It can be used in cases when the partners can not communicate with
each other in on-line way. The message is stored in a queue until it
can be processed.

Transaction demarcation interface

It is used to determine the boundaries of the transactions.

It is based on three routines:

- tpbegin
- tpcommit
- tpabort

Tpbegin : start of a global transaction. A new transaction ID is

generated. It will be attached to every activity during the
transaction. The Transaction Manager interacts with the
participating resource managers to complete the transaction
commitment or abort coordination protocol.
Tpcommit: commit the transaction, the result of updates is stored in
the database.
TpAbort: abort the transaction, the updates should be rolled back.
The main standards for open transaction processing

- X/Open XA distributed transaction processing (DTP)

which is an architecture that allows multiple programs to share
resources (e.g., databases and files) provided by multiple resource
managers and allows their work to be coordinated. The architecture
defines application programming interfaces and interactions among
transactional applications, transaction managers, resource
managers, and communications resource managers. The transaction
manager and the resource manager communicate by means of the
XA interface.

- X/Open transactional remote procedure call (TxRPC):

which allows an application to invoke local and remote resource
managers as if they were all local. TxRPC also allows an
application to be decomposed into client and server components
on different computers interconnected by means of remote
procedure calls (RPCs).