Transcript W8-4 Performance Factorsx

Software Architecture
in Practice
Architecture and Performance Factors
Introduction
• In Performance Engineering it is important to
understand the factors of the Architecture that
determines performance of your solution.
• In this presentation the most common factors are
presented.
• A basic understanding of queuing theory is useful
for understanding the impact these factors can
have on performance.
Performance Factors
• In the following slides we will look at some
factors layer by layer. Since we cannot consider
everything we will use the architecture of the
TM16 application as basis
–
–
–
–
–
–
Hardware
Operating system
Application Server
Application
Database
Network
Hardware
• The “big four”
– Processor
• no. or processors, no. of cores, hyper-threading, processor speed, cache
– Memory
• Amount of physical memory, paging, memory speed
– Disk I/O
• Disk speed, RAID configuration, local disk vs. network disk
– Network I/O
• Network interface speed, protocol, network bandwidth
• A good explanation of how to monitor these under windows can be
found here: http://www.windowsitpro.com/article/monitoring-andanalysis/monitor-windows-server-with-performance-counters
• Performance “hotspots” are typically detected from utilization and
queue lengths
– E.g. processor utilization, disk queue length etc.
Operating System
• Depending on the solution there is typically a number of configuration
options to be made, e.g.
– Size of virtual memory, network buffer sizes, number of available TCP/IP
ports, etc.
• Operating system services can also impact performance
– Unnecessary ones should be disabled
– Necessary ones should be scheduled to cause minimum impact on your
application
• Eg. run outside office hours or give lower priority
• Virtualization
– In later years virtualized servers has become more common so majority
of hosted solutions are now virtualized
– In general virtualization systems do not constrain performance
significantly, but you should know when you run virtualized – since you
might not have the capacity you think
Application Server
• An application server is a server that provides software applications
with services such as security, data services, transaction support,
load balancing, and management of large distributed systems.
– The term is often used for web servers that support the Java Platform,
Enterprise Edition, however its use isn't restricted to Java.
• Performance tuning generally starts with the Java Virtual Machine
(JVM), which serves as the foundation for the application server.
• Depending on the services used by your application you will tune
–
–
–
–
–
–
Number of Application Server instances (JVM’s)
JVM heap size
Thread pool size
Connection pool size
Caches
….
Application
• The application design has a major impact on the
performance of your system.
• In general you should focus on minimizing the work
involved in moving data from database to the user
– Don’t have too many layers
– Client must not be too chatty with the server
– Don’t fetch more data than necessary from database – and don’t
visit other components when not needed
• We will look more into patterns and anti-patterns in a later
session
Database
• Most applications use a relational database for data
storage. The database is a very common cause of
performance issues which are typically rooted in two
different aspects
– The configuration/structure of the RDBMS and the database
– The applications’ use of the database
The query optimizer is an
important piece of every DBMS.
It determines how your SQL
expression will fetch the data
you ask for.
Database optimization
• Database optimization is a complex task beyond the scope of this
class. Knowing about the basics is however a critical skill for
performance engineers.
• You always need to consider these aspects:
– Keep database statistics up-to-date
– Build appropriate indexes
– Do the work on data as close to the data as possible
• Consider stored procedures for large volumes of data
– Only retrieve the data needed from the database
• SELECT only the columns and rows needed
• Use WHERE clauses to filter data
– Use parameterized queries
• SELECT x from user where name=‘?’
– Partition large databases
… note
• And quite a few of these rules do not apply to
NoSQL databases …
CS@AU
Henrik Bærbak Christensen
10
Network
• There is typically a number of networks involved between end users
and the components of your application
• For any network the two primary factors to consider are
– Latency
– Transmission speed
• These are impacted by a number of factors, e.g.
–
–
–
–
–
–
–
–
Protocol
Number of “hops” from end-2-end
Network utilization
Network security (e.g. SSL)
Amount of messages exchanged
Message sizes
Network packet sizes
….
A few words on Java
• When using the Java programming language there are a number of
subjects you should be aware of
–
–
–
–
–
Garbage collector algorithm
Heap sizes
Temporary objects creations
Object retention leaks
….
• A good way to investigate the inner workings of your Java application
– and to learn more about general JVM performance – is to use a
java code profiler
– JDK contains a powerful graphical profiler jvisualvm
• There are a wealth of good guides on the internet describing JVM
performance, e.g.
– http://www.javaworld.com/javaworld/jw-08-2012/120821-jvmperformance-optimization-overview.html