SERVER INFRASTRUCTURE

INFRASTRUKTURA E SERVEREVE

Osman Osmani

Prishtinë
SERVER INFRASTRUCTURE
1. The definition of IT infrastructure, Cloud computing and infrastructure model
2. Availability Concepts
3. Performance Concepts
4. Security Concepts
5. Datacenters
6. Networking
7. Storage
8. Compute
9. Operating Systems
10. Infrastructure deployment options
11. Automation
12. Documenting the infrastructure, Maintaining the infrastructure
INTRODUCTION

 Performance is a typical hygiene factor

 Nobody notices a highly performing
system
 But when a system is not performing well
enough, users quickly start complaining
PERCEIVED PERFORMANCE

 Perceived performance refers to how quickly a system appears to perform its task
 In general, people tend to overestimate their own patience
 People tend to value predictability in performance
 When the performance of a system is fluctuating, users remember a bad experience
 Even if the fluctuation is relatively rare
PERCEIVED PERFORMANCE

 Inform the user about how long a task will take

 Progress bars
 Splash screens
PERFORMANCE DURING INFRASTRUCTURE
DESIGN

 A solution must be designed, implemented, and supported to meet the performance

requirements
 Even under increasing load
 Calculating performance of a system in the design phase is:
 Extremely difficult
 Very unreliable
PERFORMANCE DURING INFRASTRUCTURE
DESIGN

 Performance must be considered:

 When the system works as expected
 When the system is in a special state, like:
 Failing parts
 Maintenance state
 Performing backup
 Running batch jobs

 Some ways to do this are:

 Benchmarking
 Using vendor experience
 Prototyping
 User Profiling
BENCHMARKING

 A benchmark uses a specific test program to assess the relative performance of an

infrastructure component
 Benchmarks compare:
 Performance of various subsystems
 Across different system architectures
BENCHMARKING

 Benchmarks comparing the raw speed of parts of an infrastructure

 Like the speed difference between processors or between disk drives
 Not taking into account the typical usage of such components
 Examples:
 Floating Point Operations Per Second – FLOPS
 Million Instructions Per Second – MIPS
VENDOR EXPERIENCE

 The best way to determine the performance of a system in the design phase: use the
experience of vendors
 They have a lot of experience running their products in various infrastructure
configurations
 Vendors can provide:
 Tools
 Figures
 Best practices
PROTOTYPING

 Also known as proof of concept (PoC)

 Prototypes measure the performance of a system at an early stage
 Building prototypes:
 Hiring equipment from suppliers
 Using datacenter capacity at a vendor’s premise
 Using cloud computing resources
 Focus on those parts of the system that pose the highest risk, as early as possible in
the design process
USER PROFILING

 Predict the load a new software system will pose on the infrastructure before the
software is actually built
 Get a good indication of the expected usage of the system
 Steps:
 Define a number of typical user groups (personas)
 Create a list of tasks personas will perform on the new system
 Decompose tasks to infrastructure actions
 Estimate the load per infrastructure action
 Calculate the total load
USER PROFILING PERSONAS/TASKS

Persona Number System task Infrastructure load Frequency

of users as a result of the
per system task
persona
Data 100 Start Read 100 MB data Once a day
entry application from SAN
officer
Data 100 Start Transport 100 MB Once a day
entry application data to workstation
officer
Data 100 Enter new Transport 50 KB data 40 per
entry data from workstation to hour
officer server
Data 100 Enter new Store 50 KB data to 40 per
entry data SAN hour
officer
Data 100 Change Read 50 KB data 10 per
entry existing data from SAN hour
officer
USER PROFILING INFRASTRUCTURE LOAD

Per
Infrastructure load Per day
second
Data transport from server to workstation (KB) 10,400,000 361.1
Data transport from workstation to server (KB) 2,050,000 71.2
Data read from SAN (KB) 10,400,000 361.1
Data written to SAN (KB) 2,050,000 71.2
SCALABLE CLOUD ENVIRONMENTS

 In cloud environments, there is often no need to guess the required capacity

 Simply start with a large system and scale down if the system is underutilized. Or
vice versa: Start with a small system and scale up until the performance is
acceptable
 Cloud environments usually have very extensive logging and monitoring
capabilities to help with this
MANAGING BOTTLENECKS

 The performance of a system is based on:

 The performance of all its components
 The interoperability of various components
 A component causing the system to reach some limit is referred to as the bottleneck
of the system
 Every system has at least one bottleneck that limits its performance
 If the bottleneck does not negatively influence performance of the complete system
under the highest expected load, it is OK
PERFORMANCE TESTING

 Load testing - shows how a system performs under the expected load
 Stress testing - shows how a system reacts when it is under extreme load
 Endurance testing - shows how a system behaves when it is used at the expected
load for a long period of time
PERFORMANCE TESTING - BREAKPOINT

 Ramp up the load

 Start with a small number of virtual users
 Increase the number over a period of time
 The test result shows how the
performance varies with the load, given
as number of users versus response time.
 A cloud environment is ideal for setting
up performance testing environments
 It can be scaled up to provide the required
load and then scaled down after the test is
complete
 It can reduce the cost of running a
performance test while simulating a very
large number of users during the test period
PERFORMANCE TESTING

 Performance testing software typically uses:

 One or more servers to act as injectors
 Each emulating a number of users
 Each running a sequence of interactions
 A test conductor
 Coordinating tasks
 Gathering metrics from each of the injectors
 Collecting performance data for reporting purposes
PERFORMANCE TESTING

 Performance testing should be done in a production-like environment

 Performance tests in a development environment usually lead to results that are highly
unreliable
 Even when underpowered test systems perform well enough to get good test results, the
faster production system could show performance issues that did not occur in the tests
 To reduce cost:
 Use a temporary (hired) test environment
INCREASING PERFORMANCE ON UPPER
LAYERS

 80% of the performance issues are due to badly behaving applications

 Application performance can benefit from:
 Database and application tuning
 Prioritizing tasks
 Working from memory as much as possible (as opposed to working with data on disk)
 Making good use of queues and schedulers
 Typically more effective than adding compute power
CACHING

Time it takes to fetch 1 MB of

Component
data (ms)
Network, 1 Gbit/s 675
Hard disk, 15k rpm, 4 KB disk blocks 105
Main memory DDR3 RAM 0.2
CPU L1 cache 0.016
DISK CACHING

 Disks are mechanical devices that are slow by nature

 Caching can be implemented i:
 Disks
 Disk controllers
 Operating system
 All non-used memory in operating systems is used for disk cache
 Over time, all memory gets filled with previously stored disk requests and prefetched disk blocks,
speeding up applications.

 Cache memory:
 Stores all data recently read from disk
 Stores some of the disk blocks following the recently read disk blocks
WEB PROXIES

 When users browse the internet, data can be cached in a web proxy server
 A web proxy server is a type of cache
 Earlier accessed data can be fetched from cache, instead of from the internet
 Benefits:
 Users get their data faster
 All other users are provided more bandwidth to the internet, as the data does not have to
be downloaded again
OPERATIONAL DATA STORE

 An Operational Data Store (ODS) is a read-only replica of a part of a database, for a

specific use
 Frequently used information is retrieved from a small ODS database
 The main database is used less for retrieving information
 The performance of the main database is not degraded
FRONT-END SERVERS

 Front-end servers serve data to end users

 Typically web servers
 To increase performance, store static data on the front-end servers
 Pictures are a good candidate
 Significantly lowers the amount of traffic to back-end systems
 In addition, a reverse proxy can be used
 Automatically cache most requested data
IN-MEMORY DATABASES

 In special circumstances, entire databases can be run from memory instead of from
disk
 In-memory databases are used in situations where performance is crucial
 Real-time SCADA systems
 High performance online transaction processing (OLTP) systems
 As an example, in 2011 SAP AG introduced HANA, an in-memory database for SAP systems

 Special arrangements must be made to ensure data is not lost when a power failure
occurs
EDGE SERVERS

 The major cloud providers have datacenters around the world

 In addition, they often offer edge locations. Edge locations can be used to cache
data in close proximity to end users
 For example, a Web site that offers streaming videos can place copies of those
videos in a number of edge locations around the world to ensure a good user
experience for users in all locations
SCALABILITY

 Scalability indicates the ease in with which a system can be modified, or

components can be added, to handle increasing load
 Two ways to scale a system:
 Vertical scaling (scale up) - adding resources to a single component
 Horizontal scaling (scale out) - adding more components to the infrastructure
SCALABILITY – VERTICAL SCALING

 Adding more resources, for example:

 Server: more memory, CPU’s
 Network switch: adding more ports
 Storage: Replace small disks by larger disks
 Vertical scaling is easy to do
 It quickly reaches a limit
 The infrastructure component is “full”
SCALABILITY – HORIZONTAL SCALING

 Adding more components to the infrastructure, for example:

 Adding servers to a web server farm
 Adding disk cabinets to a storage system
 In theory, horizontal scaling scales much better
 Be aware of bottlenecks
 Doubling the number of components does not necessarily double the performance
 Horizontal scaling is the basis for cloud computing
 Applications must be aware of scaling infrastructure components
SCALABILITY – HORIZONTAL SCALING
LOAD BALANCING

 Load balancing uses multiple servers that perform identical tasks

 Examples:
 Web server farm
 Mail server farm
 FTP (File Transfer Protocol) server farm

 A load balancer spreads the load over the available machines

 Checks the current load on each server in the farm
 Sends incoming requests to the least busy server

LOAD BALANCING
LOAD BALANCING

 Advanced load balancers can spread the load based on:

 The number of connections a server has
 The measured response time of a server
 The application running on a load balanced system must be able to cope with the
fact that each request can be handled by a different server
 The load balancer should contain the states of the application
 The load balancing mechanism can arrange that a user’s session is always connected to
the same server
 If a server in the server farm goes down, its session information becomes inaccessible and
sessions are lost
LOAD BALANCING

 A load balancer increases availability

 When a server in the server farm is unavailable, the load balancer notices this and ensures
no requests are sent to the unavailable server until it is back online again
 The availability of the load balancer itself is very important
 Load balancers are typically setup in a failover configuration
LOAD BALANCING

 Network load balancing:

 Spread network load over multiple network connections
 Most network switches support port trunking
 Multiple Ethernet connections are combined to get a virtual Ethernet connection providing higher
throughput
 The load is balanced over the connections by the network switch

 Storage load balancing:

 Using multiple disks to spread the load of reads and writes
 Use multiple connections between servers and storage systems
HIGH PERFORMANCE CLUSTERS

 High performance clusters provide a vast amount of computing power by

combining many computer systems
 A large number of cheap off the-shelf servers can create one large supercomputer
 Used for calculation-intensive systems
 Weather forecasts
 Geological research
 Nuclear research
 Pharmaceutical research
 TOP500.org
GRID COMPUTING

 A computer grid is a high performance cluster that consists of systems that are
spread geographically
 The limited bandwidth is the bottleneck
 Examples:
 SETI@HOME
 CERN LHC Computing Grid (140 computing centers in 35 countries)
 Broker firms exist for commercial exploitation of grids
 Security is a concern when computers in the grid are not under control
DESIGN FOR USE

 Performance critical applications should be designed as such

 Tips:
 Know what the system will be used for
 A large data warehouse needs a different infrastructure design than an online transaction
processing system or a web application
 Interactive systems are different than batch oriented systems
 When possible, try to spread the load of the system over the available time
DESIGN FOR USE

 In some cases, special products must be used for certain systems

 Real-time operating systems
 In-memory databases
 Specially designed file systems
 Use standard implementation plans that are proven in practice
 Follow the vendor's recommended implementation
 Have the vendors check the design you created
 Move rarely used data from the main systems to other systems
 Moving old data to a large historical database can speed up a smaller sized database
CAPACITY MANAGEMENT

 Capacity management guarantees high performance of a system in the long term

 To ensure performance stays within acceptable limits, performance must be
monitored
 Trend analyses can be used to predict performance degradation
 Anticipate on business changes (like forthcoming marketing campaigns)