ch20 Database System Architecture by Rizvee

Database System Concepts, 7th
Ed.
©Silberschatz, Korth and Sudarshan
See www.db-book.com for conditions on re-
use
Chapter 20: Database System Architectures

©Silberschatz, Korth and
20.2
Database System Concepts - 7th
Outline
▪ Centralized Database Systems
▪ Server System Architectures
▪ Parallel Systems
▪ Distributed Systems
▪ Network Types

20.3
Centralized Database Systems
▪ Run on a single computer system
▪ Single-user system
▪ Multi-user systems also known as server systems.
• Service requests received from client systems
• Multi-core systems with coarse-grained parallelism
▪ Typically, a few to tens of processor cores
▪ In contrast, fine-grained parallelism uses very large number of
computers or nodes [Not belong to centralized database systems,
rather distributed database systems]
coarse-grained parallelism Few big tasks across few (4–64) CPU cores
fine-grained parallelism Many tiny tasks across many machines or processors
centralized database system is that all data is stored,
managed, and processed on a single centralized server or
system
Horizontal scaling (Scale out) : increasing the number of machines/nodes in the
cluster
Vertical scaling (Scale Up): increasing the number of CPUs, RAM, Storage

20.4
Server System Architecture
▪ Server systems can be broadly categorized into two kinds:
• Transaction servers
▪ Widely used in relational database systems, and
▪ Ensure ACID Properties: Consistency, Atomicity, Isolation, and
Durability.
• Data servers
▪ Parallel data servers used to implement high-performance
transaction processing systems

20.5
Transaction Servers
▪ Also called query server systems or SQL server systems
• Clients send requests to the server
• Transactions are executed at the server
• Results are shipped back to the client.
▪ Requests are specified in SQL and communicated to the server through a remote procedure
call (RPC) mechanism.
• SQL: SELECT * FROM Orders WHERE CustomerID = 5;
• RPC: Call executeSQL("SELECT * FROM Orders WHERE CustomerID = 5")
▪ Transactional RPC allows many RPC calls to form a transaction. So any of the failing or RPC call
will revert back all the previous calls.
1. StartTransaction(), 2. RPC: Insert new order, 3. RPC: Update customer
balance 4. RPC: Deduct inventory, 5.CommitTransaction()
▪ Applications typically use ODBC/JDBC APIs to communicate with transaction servers
Connection conn = DriverManager.getConnection(url, user, pass);
conn.setAutoCommit(false);
Statement stmt = conn.createStatement();
stmt.executeUpdate("UPDATE accounts SET balance = balance - 100 WHERE id = 1");
stmt.executeUpdate("UPDATE accounts SET balance = balance + 100 WHERE id = 2");
conn.commit();
Example with JDBC

20.6
Transaction System Processes (Cont.)

20.7
Transaction Server Process Structure
▪ A typical transaction server consists of multiple processes accessing
data in shared memory
▪ Shared memory contains shared data
• Buffer pool [For buffering]
• Lock table [Lock Mechanism]
• Log buffer [Activity Logging]
• Cached query plans (reused if same query submitted again)
▪ All database processes can access shared memory
▪ Server processes
• These receive user queries (transactions), execute them and send
results back
• Processes may be multithreaded, allowing a single process to
execute several user queries concurrently
• Typically, multiple multithreaded server processes

20.8
Transaction Server Processes (Cont.)
▪ Database writer process
• Output modified buffer blocks to disks continually
▪ Log writer process
• Server processes simply add log records to log record buffer
• Log writer process outputs log records to stable storage.
▪ Checkpoint process
• Performs periodic checkpoints
▪ Process monitor process
• Monitors other processes, and takes recovery actions if any of the
other processes fail
▪ E.g. aborting any transactions being executed by a server
process and restarting it

20.9
Transaction System Processes (Cont.)
▪ Lock manager process
• To avoid overhead of interprocess communication for lock
request/grant, each database process operates directly on the
lock table
▪ Instead of sending requests to lock manager process
• Lock manager process still used for deadlock detection
▪ To ensure that no two processes are accessing the same data
structure at the same time, databases systems implement
mutual exclusion using either
• Atomic instructions
▪ Test-And-Set
▪ Compare-And-Swap (CAS)
• Operating system semaphores
▪ Higher overhead than atomic instructions
▪ Important for process synchronization and mutual
exclusion
a semaphore is a synchronization primitive used to
control access to shared resources in a concurrent
system like multitasking processes or threads.
wait() (also called P() or down())
signal() (also called V() or up())

20.10
Atomic Instructions
▪ Test-And-Set(M)
• Memory location M, initially 0
• Test-and-set(M) sets M to 1,
and returns old value of M
▪ Return value 0 indicates
process has acquired the
mutex
▪ Return value 1 indicates
someone is already
holding the mutex
• Must try again later
▪ Release of mutex done by
setting M = 0
// Lock variable
int M = 0;
// Process trying to acquire the
lock
while (TestAndSet(M) == 1) {
// Busy-wait (spinlock)
}
// Critical section
// ...
// Release the lock
M = 0;

20.11
Atomic Instructions
▪ Compare-and-swap(M, V1, V2)
• Atomically do following
▪ If M = V1, set M = V2 and return success
▪ Else return failure
• With M = 0 initially, CAS(M, 0, 1) equivalent to test-and-set(M)
• Can use CAS(M, 0, id) where id = thread-id or process-id to record who
has the mutex

20.12
Data Servers/Data Storage Systems
▪ Data items are shipped to clients where processing is performed. Instead of
doing heavy computation or query processing on the server, the raw data is sent to the
client, which does the necessary processing (e.g., filtering, aggregating).This system is
known as client-centric system. E.g., Edge Computing.
▪ Updated data items written back to server. [Persistency]
▪ Earlier generation data servers operated on disk pages, each potentially
containing multiple data items, or occasionally at the level of
individual items. In contrast, current-generation data storage systems
operate exclusively at the level of individual data items.
▪ Current generation commonly includes:
• Commonly used data item formats include JSON, XML, or just
uninterpreted binary strings

20.13
Data Servers/Storage Systems (Cont.)
▪ Prefetching
• Prefetch items that may be used soon
▪ Data caching
• Cache coherence
▪ Lock caching
• Locks can be cached by client across transactions
• Locks can be called back by the server
▪ Adaptive lock granularity
• Lock granularity escalation
▪ switch from finer granularity (e.g. tuple) lock to coarser
• Lock granularity de-escalation
▪ Start with coarse granularity to reduce overheads, switch to
finer granularity in case of more concurrency conflict at server
▪ Details in book

20.14
Data Servers (Cont.)
▪ Data Caching
• Data can be cached at client even in between transactions
• But check that data is up-to-date before it is used (cache coherency)
• Check can be done when requesting lock on data item
▪ Lock Caching (not immediate release of the lock after transaction
execution)
• Locks can be retained by client system even in between transactions
• Transactions can acquire cached locks locally, without contacting server
• Server calls back locks from clients when it receives conflicting lock request.
If another client requests a lock that conflicts (e.g.,
wants to write to the same data), the server sends a
callback to the caching client: return the lock
▪ If any local transaction is using it, it can not return,
otherwise it returns
▪ Client returns lock once no local transaction is using it.
▪ This idea is similar to lock callback on prefetch, but across transactions,
keeping lock though the transaction has already ended.

20.15
Parallel Systems
▪ Parallel database systems consist of multiple processors and
multiple disks connected by a fast interconnection network.
▪ Motivation: handle workloads beyond what a single computer
system can handle
▪ High performance transaction processing
• E.g. handling user requests at web-scale
▪ Decision support on very large amounts of data
• E.g. data gathered by large web sites/apps

20.16
Parallel Systems (Cont.)
▪ A coarse-grain parallel machine consists of a small number of
powerful processors
▪ A massively parallel or fine grain parallel machine utilizes thousands
of smaller processors.
• Typically hosted in a data center
▪ Two main performance measures:
• Throughput --- the number of tasks that can be completed in a
given time interval
• Response time --- the amount of time it takes to complete a single
task from the time it is submitted

20.17
Speed-Up and Scale-Up
▪ Speedup: a fixed-sized problem executing on a small system is given
to a system which is N-times larger.
• Measured by:
speedup = small system elapsed time
large system elapsed time
• Speedup is linear if equation equals N.
▪ Scaleup: increase the size of both the problem and the system
• N-times larger system used to perform N-times larger job
• Measured by:
scaleup = small system small problem elapsed time
big system big problem elapsed time
• Scale up is linear if equation equals 1.

20.18
Speedup
A particular problem that does not speed
up even if the presence of additional
resources

20.19
Scaleup
TS = Time scale for system coordination, communication,
overhead
TL = Time scale for local computation or problem solving

20.20
Batch and Transaction Scaleup
▪ Batch scaleup:
• A single large job; typical of most decision support queries and
scientific simulation.
• Use an N-times larger computer on N-times larger problem.
▪ Transaction scaleup:
• Numerous small queries submitted by independent users to a
shared database; typical transaction processing and timesharing
systems.
• N-times as many users submitting requests (hence, N-times as
many requests) to an N-times larger database, on an N-times
larger computer.
• Well-suited to parallel execution.

20.21
Factors Limiting Speedup and Scaleup
Speedup and scaleup are often sublinear due to:
▪ Startup/sequential costs: Cost of starting up multiple processes,
and sequential computation before/after parallel computation
• May dominate computation time, if the degree of parallelism is
high
• Suppose p fraction of computation can be parallelized. So, (1-p)
fraction of the computation must be run sequentially.
• Amdahl’s law: speedup limited to: 1 / [(1-p)+(p/n)]
• Gustafson’s law: scaleup limited to: (1-p) + np
▪ Interference: Processes accessing shared resources (e.g.,system bus,
disks, or locks) compete with each other, thus spending time waiting
on other processes, rather than performing useful work.
▪ Skew: Increasing the degree of parallelism increases the variance in
service times of parallely executing tasks. Overall execution time
determined by slowest of parallely executing tasks.

20.22
Amdahl’s law: speedup limited to: 1 / [(1-p)+(p/n)]
Purpose: Amdahl's Law calculates the maximum theoretical speedup of a
program when increasing the number of processors for a fixed problem size. It
assumes that the total amount of work remains constant regardless of the
number of processors.
Total Time with Parallelism:
● New execution time = (Time for serial part) + (Time for parallel part on n
processors)
● New execution time = (1 p)×1+(p)×(1/n)
−
● New execution time = (1 p)+(p/n)
−
Speedup Calculation: Speedup is defined as:
● S=Original Execution Time/New Execution Time
● Since we normalized original time to 1,
● S=1/[(1 p)+(p/n)]
−

20.23
Gustafson’s law: scaleup limited to: (1-p) + np
Purpose: Gustafson's Law calculates the scaled speedup for a parallel system where the problem size scales with
the number of processors. It addresses the limitation of Amdahl's Law by assuming that as you add more
processors, you also solve a larger problem in roughly the same amount of time.
Workload Scales with Processors: The core idea here is that if you have more processors, you'll tackle a
proportionally larger problem.
● Assume that the parallel part of the program scales linearly with n processors.
● Let the time taken by the parallel portion on a single processor be p.
● Let the time taken by the serial portion on a single processor be (1−p).
● The total time on a single processor for the scaled problem would be (1−p)+n×p (because the parallel part is
now n times larger for the n processors).
Time on n Processors: On n processors, the serial part still takes (1−p) time. The scaled parallel part, which took
n×p on a single processor, now takes just p time on n processors (since it's distributed among them).
● Total time on n processors = (1−p)+p=1. (This is the key assumption of Gustafson's Law: you scale the problem
size such that the execution time on n processors is approximately the same as the original problem on 1
processor).
Speedup Calculation: Speedup is defined as:
● S=Original Execution Time (scaled problem on 1 processor)/New Execution Time (scaled problem on n
processors)
● S=[(1−p)+np]/1
● S=(1−p)+np

20.24
Interconnection Network Architectures
▪ Bus. System components send data on and receive data from a single
communication bus;
• Does not scale well with increasing parallelism.
▪ Mesh. Components are arranged as nodes in a grid, and each
component is connected to all adjacent components
• Communication links grow with growing number of components,
and so scales better.
• But may require 2√n hops to send message to a node (or √n with
wraparound connections at edge of grid).
▪ Hypercube. Components are numbered in binary; components are
connected to one another if their binary representations differ in
exactly one bit.
• n components are connected to log(n) other components and can
reach each other via at most log(n) links; reduces communication
delays.
▪ Tree-like Topology. Widely used in data centers today

20.25
Interconnection Architectures

20.26
Interconnection Network Architectures
▪ Tree-like or Fat-Tree Topology: widely used in data centers today
• Top of rack switch for approx 40 machines in rack
• Each top of rack switch connected to multiple aggregation switches.
• Aggregation switches connect to multiple core switches.
▪ Data center fabric

20.27
Network Technologies
▪ Ethernet
• 1 Gbps and 10 Gbps common, 40 Gbps and 100 Gbps are
available at higher cost
▪ Fiber Channel
• 32-138 Gbps available
▪ Infiniband
• A very-low-latency networking technology
▪ 0.5 to 0.7 microseconds, compared to a few microseconds
for optimized ethernet

20.28
Parallel Database Architectures

20.29
Shared Memory
▪ Processors (or processor cores) and disks
have access to a common memory
• Via a bus in earlier days, through an
interconnection network today
▪ Extremely efficient communication
between processors
▪ Downside: shared-memory
architecture is not scalable beyond 64
to 128 processor cores
• Memory interconnection network
becomes a bottleneck

20.30
Modern Shared Memory Architecture
Own local memory
However, through X or
interconnection any processor
can share other’s memory

20.31
Cache Levels
▪ Cache line: typically 64 bytes in today’s processors, the smallest unit of data
which i s transferred between RAM (main memory) and CPU caches
▪ Cache levels within a single multi-core processor
▪ Shared memory system can have multiple processors, each with its own
cache levels

20.32
Cache Coherency
▪ Cache coherency:
• Local cache may have out of date value
• Strong vs weak consistency models
• With weak consistency, need special instructions to ensure cache is
up to date
▪ Memory barrier instructions
• Store barrier (sfence)
▪ Instruction returns after forcing cached data to be written to
memory and invalidations sent to all caches. An sfence instruction
ensures that all write (store) operations initiated before the sfence
instruction are completed and made globally visible
• Load barrier (lfence)
▪ Returns after ensuring all pending cache invalidations are
processed. An lfence instruction ensures that all read (load) operations initiated before the
lfence instruction are completed before any read operations initiated after the lfence instruction
• mfence instruction does both of above
▪ Locking code usually takes care of barrier instructions
• Lfence done after lock acquisition and sfence done before lock
release
Buffer -> cache -> main
memory -> disk

20.33
Shared Disk
▪ All processors can directly access all
disks via an interconnection
network, but the processors have
private memories.
• Architecture provides a degree
of fault-tolerance — if a
processor fails, the other
processors can take over its
tasks
▪ the data of the failed
processor is resident on
disks that are accessible
from all processors.
▪ Downside: bottleneck now occurs at
interconnection to the disk
subsystem. Too much processor / too much
disk connectivity

20.34
Storage Area Network (SAN)

20.35
Shared Nothing
▪ Node consists of a processor,
memory, and one or more disks
▪ All communication via
interconnection network
▪ Can be scaled up to thousands of
processors without interference.
▪ Main drawback: cost of
communication and non-local disk
access; sending data involves
software interaction at both ends.

20.36
Hierarchical
▪ Combines characteristics of shared-memory, shared-disk, and shared-
nothing architectures.
• Top level is a shared-nothing architecture
▪ With each node of the system being a shared-memory system
• Alternatively, top level could be a shared-disk system
▪ With each node of the system being a shared-memory system

20.37
Shared-Memory Vs Shared-Nothing
▪ Shared-memory internally looks like shared-nothing!
• Each processor has direct access to its own memory, and indirect
(hardware level) access to rest of memory
• Also called non-uniform memory architecture (NUMA)
▪ Shared-nothing can be made to look like shared memory
• Reduce the complexity of programming such systems by
distributed virtual-memory abstraction
• Remote Direct Memory Access (RDMA) provides very low-latency
shared memory abstraction on shared-nothing systems
▪ Often implemented on top of infiniband due it its very-low-
latency
• But careless programming can lead to performance issues

20.38
Distributed Systems
▪ Data spread over multiple machines (also referred to as sites or
nodes).
▪ Local-area networks (LANs)
▪ Wide-area networks (WANs)
• Higher latency

20.39
Distributed Databases
▪ Homogeneous distributed databases
• Same software/schema on all sites, data may be partitioned
among sites
• Goal: provide a view of a single database, hiding details of
distribution
▪ Heterogeneous distributed databases
• Different software/schema on different sites
• Goal: integrate existing databases to provide useful functionality
▪ Differentiate between local transactions and global transactions
• A local transaction accesses data in the single site at which the
transaction was initiated.
• A global transaction either accesses data in a site different from
the one at which the transaction was initiated or accesses data in
several different sites.

20.40
Data Integration and Distributed Databases
▪ Data integration between multiple distributed databases
▪ Benefits:
• Sharing data – users at one site able to access the data residing at
some other sites.
Example:
● A university has a distributed database: one part stores student
records, another stores financial information, and a third stores
library data, each on different servers or campuses.
● A student advisor can pull data from all three databases through a
single query interface — even though the data is stored on
different systems.
• Autonomy – each site is able to retain a degree of control over data
stored locally.
▪ In a supply chain network, each company (manufacturer,
distributor, retailer) maintains its own database.
▪ They integrate their systems for inventory tracking or order
management, but each company still decides how its database is
structured, who can access it, and how updates are made.

20.41
Availability
▪ Network Partitioning: A network partition occurs when communication
between parts of a distributed system is disrupted — nodes are split
into isolated "partitions" that can't talk to each other, even though
they may still be running.
▪ Availability of system
• If all nodes are required for system to function, failure of even one
node stops system functioning.
• Higher system availability through redundancy
▪ data can be replicated at remote sites, and system can function
even if a site fails.

20.42
Implementation Issues for Distributed
Databases
▪ Atomicity needed even for transactions that update data at multiple
sites
▪ The two-phase commit protocol (2PC) is used to ensure atomicity
• Basic idea: each site executes transaction until just before commit,
and the leaves final decision to a coordinator
• Each site must follow decision of coordinator, even if there is a
failure while waiting for coordinators decision
▪ 2PC is not always appropriate: other transaction models based on
persistent messaging, and workflows, are also used
▪ Distributed concurrency control (and deadlock detection) required
▪ Data items may be replicated to improve data availability
▪ Details of all above in Chapter 24

20.43
Cloud Based Services
▪ Cloud computing widely adopted today
• On-demand provisioning and elasticity (Memory increase,
processor increase)
▪ ability to scale up at short notice and to release of unused
resources for use by others
▪ Infrastructure as a service
• Virtual machines/real machines
▪ Platform as a service
• Storage, databases, application server
▪ Software as a service
• Enterprise applications, emails, shared documents, etc,
▪ Potential drawbacks
• Security
• Network bandwidth

20.44
Cloud Service Models

20.45
Application Deployment Alternatives
Individual Machines Virtual Machines Containers
(e.g. VMWare, KVM, ..) (e.g. Docker)

20.46
Application Deployment Architectures
▪ Services
▪ Microservice Architecture
• Application uses a variety of services
• Service can add or remove instances as required
▪ Kubernetes supports containers, and microservices

Database System Concepts, 7th
Ed.
©Silberschatz, Korth and Sudarshan
See www.db-book.com for conditions on re-
use
End of Chapter 20

ch20 Database System Architecture by Rizvee

ch20 Database System Architecture by Rizvee

More Related Content

Similar to ch20 Database System Architecture by Rizvee (20)

Recently uploaded (20)

ch20 Database System Architecture by Rizvee