Unit-IV

PERFORMANCE AND TUNING

1 PERFORMANCE AND TUNING
What is performance and tuning process?
Performance tuning is primarily a matter of resource management and correct
system-parameter setting. Tuning the workload and the system for efficient resource use
consists of the Identifying the workloads on the system. Setting objectives: Determining
how the results will be measured.
Performance and tuning is a methodology to maximize throughput (the amount of
a product or service that a company can produce and deliver to a client within a specified
period of time) and minimize response times of batch jobs, online transactions, and
Internet activities.
There are actually two related activities normally combined into one process. We
continually tune infrastructure hardware and software, as well as their interrelationships,
to improve the performance of systems. The tuning activity consists of a variety of
analyses, adjustments, and changes to a whole host of parameters.
The five infrastructure areas most impacted by performance and tuning are:
• Servers
• Disk storage
• Databases
• Networks
• Desktop computers
Differences between Performance & Tuning Process and Other processes

Performance and tuning Other processes

 Consists primarily of two major activities  Consists primarily of one major

 Normally has multiple sub process activity
owners  Normally has one overall process
 Shares ownership across multiple owner
departments  Centralizes ownership in a single
 Task have continuous and ongoing department
nature  Tasks have definitive start and end
 Process is highly iterative dates
 Process tools vary widely among the  Processes are rarely iterative
resource environments  Process tools are usually shared
 Process utilizes a large number of across departments
differentiated metrics(identify risk areas,  Processes utilizes a small number of
priorities initiatives, set targets) similar metrices
Solution for performance problem
 The quickest and simplest way to solve a performance problem is to throw more
hardware at it. For Example
 The system is running slow, so upgrade the processors
 If swap rates are too high, just add more memory;
 For cache hits too low or disk extents too high, simply buy more cache and
disk volumes
 The tuning is an ongoing activity in which the performance bottleneck is never
truly eliminated; it’s only minimized or relocated.
 Example: Hundreds of new desktop computers have been added to a highly
used application.
 The resulting slow response is attributed to lack of adequate bandwidth on
the network, so bandwidth is added.
2 Preferred Characteristics of a Performance and Tuning Process
Owner
 There may be two or more sub processors whose high-priority characteristics vary
depending on the area in which they are working.
 For example, an individual selected as the process owner for only the network area
should have as a high priority knowledge of network software and components,
 But knowledge of systems software and components need only be a medium
priority.
 The reverse would apply to an individual selected as the process owner for only
the server area.
Performance and Tuning Applied to the Five Major Resource
Environments
1 Server Environment
 Types and sizes of processor platforms, including mainframe computers, midrange
computers, workstations, and servers all of these platforms are together referred as
servers.
 Major performance issues in a server environment:
1. Processors
2. Main memory
3. Cache memory
4. Number and size of buffers
5. Size of swap space
6. Number and type of channels
 Processor - The number and power of processors influence the rate of work
accomplished for processor-oriented transactions.
 CPU utilization can vary according to the type and amount of computing task
because some tasks require heavy CPU time while others require less CPU time
Cache memory - It is available on most mainframe computers and on some models of
servers, to provide additional means of tuning transaction processing.
 It is slower than main memory but faster than secondary storage.
 One of the most effective ways to improve disk-storage performance is to utilize
cache memory in RAID-type disk arrays. (RAID stands for redundant array of
independent—originally, inexpensive—disks.
 The reason cache is so effective at improving disk performance is that, for an
overwhelming number of input or output transactions to and from disks, it
eliminates the time-consuming steps of seeking and searching for data out on disk
volumes.
 It accomplishes this through the use of ingenious pre-fetch algorithms that
anticipate which data will be requested next and then preload it into the high-speed
cache memory.
 A pre-fetch algorithm is a set of software instructions that analyses the patterns
of disk storage read-request activity to anticipate and pre-load, or pre-fetch,
expected records of data.
Main memory - It is extremely fast circuitry that stores instructions, data, and most of
the operating system software, all of which are likely to be used immediately by the
processor.
 The number and size of buffers -Buffers are high-speed registers of main
memory that store data being staged for input or output. The number and size of
 buffers assigned for processing of I/O operations can trade off the amount of
memory available for processor-oriented transactions.

 Number and type of channels- Channels are physical cables that connect the
main memory to external I/O devices such as disk drives, tape drives, and printers.
The rate of processing I/O operations is also determined by the number and speed
of channels connecting servers to external disk equipment.

 The size of swap space -The portion of main memory set aside(allocate) for
storing small frequently used part of programs from main memory to secondary
storage is called swap space. The size of swap space can be adjusted to match the
profiles of application and database processing.

 Performance metrics in a server environment include:

1. Processor utilization percentages (percentage of time the processor spent
doing work)
2. The frequency of swapping in and out of main memory
3. Percentage of hits to main memory cache
4. The length and duration of processing queues
5. The percentage of utilization of channels
6. The amount of processor overhead (work) used to measure performance
2 Disk Storage Environment
 The second type of infrastructure resources impacted by performance and tuning is
disk-storage equipment.
 The following list indicates the major performance issues in a disk-storage
environment:
1. Cache memory
2. Volume groups
3. Striping
4. Storage area networks
5. Network‐attached storage
6. Extents
7. Fragmentation
 Mapping out volume groups is another effective way to improve performance.
Volume groups are a logical grouping of physical disk drives. It reduces seek-and-
search times by combining frequently used groups of physical volumes into one
logical group.
Example: By arranging the data of customers using credit and debit card in a
single disk is more efficient than spreading across different disks.
 Striping is a performance-improvement technique in which long blocks of data
that are to be read or written sequentially are stored across multiple drives, usually
within a logical volume group. This is done to increase the number of data paths
 and transfer rates and to allow for the simultaneous reading and writing to multiple
disks
 A storage area network (SAN) is a configuration enhancement that places a
high-speed fibre optic switch between servers and disk arrays. It is a dedicated,
independent high-speed network that interconnects and delivers shared pools
of storage devices to multiple servers.
 Each server can access shared storage as if it were a drive directly attached to the
server. The fibre channel can transmit data between the servers and the arrays at
speeds of up to 1 or 2 terabytes per second
 A network attached storage (NAS) device is a server that is dedicated solely to
file sharing. NAS is a single storage device that serves files over Ethernet and is
relatively inexpensive and easy to set up, while a SAN is a tightly coupled
network of multiple devices that is more expensive and complex to set up and
manage.
 Extents occur when the amount of data needing to be written exceeds the amount
of original contiguous disk space allocated. Rather than abnormally terminating
the request, the operating system looks for additional available space on the disk
volume and extends the original contiguous space.
 The process of Reallocating files and compressing them to reduce extents is
known as Fragmentation It occurs when a file is broken up into pieces to fit on
the disk.
 Because files are constantly being written, deleted and resized, fragmentation is a
natural occurrence. When a file is spread out over several locations, it takes longer
to read and write.
3 Database Environment
The physical layout of a database can greatly influence the eventual response
times of inquiries or updates. Eight of the most common performance issues associated
with a database environment are shown in the following list:
1. Placement of table files
2. Initialization parameters
3. Placement of data files
4. Indexes and keys
5. Locks
6. Balance of system resources
7. Access patterns
8. Database fragmentation
A table file describes the various tables that reside within the database, is a critical
navigational map of a database that serves as its data dictionary. Most database requests
are first channelled through the table file.
The optimal location for this file is in the main memory of the server where fetch
times are the quickest. In some cases it is loaded into the cache memory of the disk array
when space is available.
 Keys are a designated field within a record of a file or a database that enables the
quick and secure searching or sorting of records within a table. Multiple keys are allowed
for a given table, to sort tables in multiple orders. Multiple constraints can be used to
retrieve data quickly with the use of keys. Keys can also be used to chain together groups
of transactions that are normally executed sequentially.
Indexes are lists of keys within a database that enable the fast searching of a
particular record or groups of records. The greater the number of indexes, overhead.
Locks protect the integrity of a record of data by making it inaccessible to
unauthorized users. Locks are used with multi-user files and databases to ensure multiple
updates are not made simultaneously. Locks are sometimes used to protect the integrity
of data records and to ensure single, sequential updates.
Balancing system resources of processors, memory, channel, and disk is
necessary to ensure no single resource becomes under-utilized by affecting good
performance. Monitoring the usage of these various resources to ensure no more
resources are under- or over-utilized.
Example, once a server approaches 80 percent utilization, action should be taken to add
capacity or, if possible, to re-distribute the workload to under-utilized servers.
Understanding the access patterns into a database is one of the best ways to
optimize an application’s performance. It includes anticipating what the read and write
sequences To which parts of the database read or write occur, Which part allows database
administrators to map out the physical database in an optimal configuration.
Database records are rewritten in locations that are different from their optimal
starting points. This causes two problems:
1. The data is no longer optimally placed for its anticipated access pattern.
2. The data records are now fragmented, causing less than efficient use of the disk
space.
Periodically defragmenting the database with software utilities corrects both problems.
When a program saves a file on a disk, it puts the file into an empty space on the disk. ...
Disk defragmentation takes all the pieces of each file, and stores them in one place.
4 Network Environment
There are a variety of factors that influence the performance of a network,
beginning with its overall design and topology. In this section we discuss some of the
more common issues:
1. Bandwidth
2. Line speed
3. Protocols
4. Single sign‐ons
5. Number of retries
6. Nonstandard interfaces
7. Broadcast storms
Network bandwidth: A robust capacity planning forecast is a good place to start a
network bandwidth calculation. It should include a reasonable forecast of the number of
concurrent users, the peak transaction traffic loads, and an estimate of transaction arrival
patterns and types. Network workload simulators can be used to test transaction response
times under varying loads with different amounts of bandwidth.
Line speed: The faster the speed, the greater the performance. Line speed refers to how
many bits per second your Internet connection can download. Just as in a server and disk
environment, resources need to be balanced and cost-justified, if lines must be upgraded
for performance reasons, it pays to compare suppliers, shop around, and negotiate for
terms such as maintenance, tariffs, and add-ons.
A network protocol is an established set of rules that determine how data is
transmitted between different devices in the same network. Essentially, it allows
connected devices to communicate with each other, regardless of any differences in their
internal processes, structure or design.
Single sign-ons: It is an authentication method that enables users to securely authenticate
with multiple applications and websites by using just one set of credentials. Here there is
a trade-off between performance and security.
The convenience and performance savings of logging on only once instead of
multiple times for access to the network, operating system, database management system,
and a particular application must be weighed against the potential security exposures of
bypassing normal password checks. Most applications have robust security features
designed into them to mitigate this risk.
Number of retries: Transmission errors occur periodically on all network lines. In most
instances, the errors are quickly detected, the network request is retried If errors are
temporary in nature the retransmission is successful.
If the retransmission is not successful, multiple retries are attempted. After a
certain number of unsuccessful retries, the error is designated as permanent and more
sophisticated diagnostics are activated. The number of retries to be attempted impacts
performance and should be monitored and adjusted appropriately.
Nonstandard interfaces and broadcast storms: Connecting devices with nonstandard
interfaces (poor technology, low port rate switches and improper network configurations)
to the network can cause major performance problems such as locking up lines,
introducing interference, or, in extreme cases, flooding the network with nonstop
transmissions called broadcast storms. Caution should be exercised and a heightened
awareness of network performance put in place whenever an exception to network
standards is permitted.
5 Desktop Computer Environment
The key issues that were discussed here are as follows
1. Processors
2. Memory
3. Disk storage space
4. Network connections
 5. Diagnostic tools
6. Administrative tools
Processor and memory: The size and power of the processing chip required by a
desktop computer is influenced by the number and type of applications that will be run.
Spreadsheets, statistical analysis programs, and highly graphical routines need much
more processing power to deliver acceptable response than word-processing applications.
The amount of processing power and memory needed to ensure acceptable
performance depends on the number and types of windows expected to be open
simultaneously. Two windows consisting of email and textual documents need far fewer
resources than six windows that include compound documents and multimedia streaming
videos.
No of Application in background that run in the background that are automatically
loaded during start-up also affects the desktop back ground environment.
New and advanced features of modern desktop operating systems like Microsoft vista
may also use large amounts of capacity.
These background applications include:
• Automatic updaters for applications
• Automatic updaters for operating systems
• Anti-virus software
• Restriction software to protect children
• Anti-spam software
• Personal firewalls
• Anti-spyware
These various applications can consume half of the processing capability of the desktop
without the user even being aware of it.
The amount of disk storage space has a less direct effect on the performance of
desktop computers compared to the processor or memory. External desktop disks impacts
performance by taking high response time for retrieving data.
Network connections: Desktop devices are attached to LANs and WANs through
network connections. The speed of the lines and the modems that utilize these network
connections impact performance at the desktop the number of desktops on the networks,
and the corresponding number of hubs and routers that interconnect the various network
segments and devices, also influences desktop performance.
Poorly initialized parameters of Diagnostic tools that enable a remote technician to
take control of a desktop computer to troubleshoot problems adversely impact
performance. Diagnostic tools also are capable of running in real-time on the desktop, if
left running, they can easily impact desktop performance.
Performance can also be impacted by administrative tools, such as those that
automatically inventory the configuration of the desktop through an asset-management
system. They can also interpret and process large amounts of detailed information about
the internal specifications of the desktop, which can slow down normal operations.