Upload
nguyentuyen
View
230
Download
1
Embed Size (px)
Citation preview
Session 40A Application Performance and How Apdex Makes it Better
Application Performance and How Apdex Makes it Better
CMG International ConferenceSan Diego, CaliforniaDecember 5, 2007
Peter SevcikNetForecast, Inc.955 Emerson DriveCharlottesville, VA 22901
[email protected] 249 1310
Session 40A
©2007, Apdex Alliance, Inc. All rights reserved. Slide 2
Apdex Symposium
Apdex Case Studies45A4:00-5:00
Setting Performance Objectives Using Apdex44A2:45-3:45
Measurement Tools and Reports43A1:15-2:15
Apdex Process41A9:15-10:15
Application Performance and How Apdex Makes it Better40A8:00-9:00
We thank the Contributing Members for their financial support of the Alliance
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 3
Outline
BSM = TSM + APMApdex
© 1990-2007 Peter Sevcik and NetForecast, Inc., All rights reserved.
©2007, Apdex Alliance, Inc. All rights reserved. Slide 4
Application Delivery Complexity
WANsAccess
Firewalls& IDS
LANSwitches& Routers
LoadBalancing
System
Servers VirtualMachines
ResourcePools
app
OS
OS
OS
Applications
Technology Silos
Single User Traffic Flow Flow Processing Point
What Matters: Task = User hits “Enter” to system “Responds”What Matters: Task = User hits “Enter” to system “Responds”
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 5
Many Columns
The datacenter now houses:Access system (firewalls, intrusion detection)LAN (many layers of Gigabit Ethernet switches)Load balancers (server directors, TCP shaping, SSL offload)Web serversApplication serversDatabase serversSAN (storage area network)Storage
However, many of the servers are in fact operating within virtual machines. So there is the added complexity of:
Virtual machinesVM resource pools
Let us not forget the:Client (any kind of machine running any kind of browser)Network (private or Internet)
©2007, Apdex Alliance, Inc. All rights reserved. Slide 6
Many Rows
Tasks: Users interact with an application one task at a timeFrom the Apdex Specification:Task time is measured from the moment the user enters an application query, command, function, etc., that requires a server response to the moment the user receives the response such that they can proceed with the application. Often called the “user wait time” or “application response time”.
Turns: Tasks comprise many turns and a lot of payloadEach application client-server software interaction needed to generate a user response or task. These software-level client-server interactions add to the time it takes for the software to complete a task. A turn is a client-server request-driven round-trip. Often called application “chattiness”.Web-based B-B application turns have grown
– 1995: 20 turns per task– 2007: 88 turns per task– Tasks are on a 13% compound annual growth rate—with no end in sight
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 7
Application Profiles Sample
1
10
100
1,000
Payload per Task (Bytes)
App
licat
ion
Turn
s pe
r Tas
k (c
ount
)
1,000 10K 100K 1M 10M
CIFS
SAPGUI
SAPWeb
MAPI
WEB2002
WEB2007
©2007, Apdex Alliance, Inc. All rights reserved. Slide 8
How Applications Are Delivered
TechnologiesResourcesSilosAssetsColumns
Many ColumnsA typical user may touch 12 subsystems to execute a task
FlowsTrafficTurns (app round trips)User ExperienceRows
Many RowsA typical task may require 10-100 turns to execute a task
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 9
Visualizing How an Application Works for One User(Rows and Columns)
12 Technologies
30 T
urns
Each column representsthe device used withineach technology
Each row representsa turn within eachtask within the flow
©2007, Apdex Alliance, Inc. All rights reserved. Slide 10
Two Views of Performance
12 Technologies
30 T
urns
PerformanceDelivered Note:
Each taskrequires10 turns,or rows
Every column deliveredthe same performance
This task hadsignificantlypoorer performancewith 4-times thenumber of reds
TaskA
TaskB
TaskC
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 11
Real World Examples
Commercial flights have strict flight rules. The FAA tracks every plane throughout the flight.
Private pilots fly in and out of general aviation terminals that are well maintained. They do not have to file a flight plan.
Airline Service
The card company keeps track of your buying habits. It calls you when there is a suspicious charge and cancels the charge.
The credit card company checks your credit, processes debits and payments, keeps a strict set of rules on how you must pay.
Credit Card Service
FedEx provides on-line 24-hour detailed tracking of your package from door-to-door.
The postal service keeps their offices open and their carriers are punctual. You post a parcel and it arrives at the other end.
Package Delivery
The lights are synchronized to maximize traffic flow and each car gets through the city faster.
Each light is operating, no roads are congested, all must be well.
City Traffic Lights
RowColumn
©2007, Apdex Alliance, Inc. All rights reserved. Slide 12
Service Delivery Frameworks
ITIL and ITSM (Columns)The IT Infrastructure Library® (ITIL) describes the organisation of IT resources to deliver business value, and documents processes, functions and roles in IT Service Management (ITSM)ITSM (IT Service Management) is focused on managing technologiesand software not user flows
Application Performance Management – APM (Rows) is missing in ITIL and ITSM
APM is the art and science of making applications run well by linking application performance goals to business objectivesApplied during ongoing operations of a business applicationAPM provides process and tools to manage user flowsAPM is required for an end-user experience SLA
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 13
ITIL Processes Related to APM
Service OperationEvent ManagementIncident ManagementRequest FulfillmentProblem ManagementAccess ManagementOperational Activities of Processes …– Change Management– Configuration Management– Release and Deployment Management– Capacity Management– Continuity Management– Availability Management– Knowledge Management– Financial Management for IT Services
Continual Service ImprovementService Level Management
Minimal set of processesthat relate to ongoingAPM delivery quality
ITIL v3 published in May 2007comprises 5 volumes:
Service StrategyService DesignService TransitionService OperationContinual Service Improvement
©2007, Apdex Alliance, Inc. All rights reserved. Slide 14
ITIL Service Delivery Applied to Flows
Incident ManagementLoss of user access, loss of service within a geographic region, slow performance, software incompatibility (client and server failing to communicate), missing cookies, can’t acquire address or credentials, etc.
Availability ManagementAll authorized user access methods (wireline, cell service, WiFi) are working, client-server connections can be made, all devices on a flow path are operating, SSL keys are installed and certified, alternate routing, etc.
Capacity ManagementSufficient bandwidth for each flow, QoS and precedence handling, load balancing, traffic control, sufficient TCP connection pools, latency within application needs, proper application acceleration techniques applied, etc.
Service Level ManagementFlow characteristics are known and supported, user response time supports the business function, voice services meet quality standards, videoconferencing supports business function, etc.
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 15
Different Reports on the Same Infrastructure
Availability
Incident
Service LevelService Level
Availability
Incident
Service Level
Availability
CapacityCapacityCapacity
Incident
Per technology, device,subsystem, system
Per flow, application,user group, location
CoreDevices
AndSystems
TechnologyService Management
ApplicationPerformance Management
Availability
Incident
Service LevelService Level
Availability
Incident
Service Level
Availability
CapacityCapacityCapacity
Incident
©2007, Apdex Alliance, Inc. All rights reserved. Slide 16
Availability
The Views Diverge With Sophistication
TechnologyService Management
ApplicationPerformance Management
Incident
Capacity
Srvc Level
Incident
Availability
Capacity
Srvc Level
CommonAttributes
UserCentric
AssetCentric
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 17
Diverging Granularity
CommonAttributes
UserCentric
AssetCentric
Utilization ofeach Device
Actual load
Predictedaggregate load
Utilization ofeach subsystem
(memory, disk, etc)
Actual traffic
Source-Destinationflows by location
Individualuser flows
(resp time, VoIP quality, etc)
Mor
e G
ranu
larit
yDetail helps
capacity planningand ROI
Detail helpslink to business
and SLA
©2007, Apdex Alliance, Inc. All rights reserved. Slide 18
Stages of Maturity
Reactive DiagnosisDiagnostic tools, problem triage, fault identification
Proactive InterventionOngoing measurement, trend analysis, resource planning
Quality WarrantyService objectives are stated
Portfolio ManagementManaging many business applications as a group
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 19
APM Value Framework
Reactive Diagnosis
Proactive Intervention
QualityWarranty
Portfolio Management
Service LevelCapacityAvailabilityIncident
©2007, Apdex Alliance, Inc. All rights reserved. Slide 20
BSM – Business Service Management
APMApplication Performance Management
Flows
TSMTechnology Service Management
Technologies
Service LevelCapacityAvail-
abilityIncident
Reactive Diagnosis
Proactive Intervention
QualityWarranty
Portfolio Management
Service LevelCapacityAvail-
abilityIncident
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 21
Stereoscopic View
Stereoscopic views of ITITSM assumes that if service is available, the user experience is satisfactoryAPM assumes that if response time is satisfactory, the service is availableEach image is convincing but each is incomplete
Good BSM needs bothITSM is a necessary but insufficient service modelAPM can’t exist without a good TSM foundation
The 3-D effect adds clarity and new information not
seen in each image alone
©2007, Apdex Alliance, Inc. All rights reserved. Slide 22
BSM Framework Supported
APMApplication Performance Management
Flows
TSMTechnology Service Management
Technologies
Service LevelCapacityAvail-
abilityIncident
Reactive Diagnosis
Proactive Intervention
QualityWarranty
Portfolio Management
Service LevelCapacityAvail-
abilityIncident
ApdexApplies
AppSLA
More Valu
e
More Valu
e
Should be balanced
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 23
Outline
BSM = TSM + APMApdex
© 1990-2007 Peter Sevcik and NetForecast, Inc., All rights reserved.
©2007, Apdex Alliance, Inc. All rights reserved. Slide 24
Today’s Problem:Many Numbers, Little Insight
Measured Response Time (seconds)
App A App B App C App D App E
Day Average 6.0 12.5 3.1 8.4 2.0
BestHour 5.0 6.8 2.8 4.1 1.7
Worst Hour 18.6 18.9 8.6 19.3 6.5
95th
Percentile 8.1 17.3 10.7 12.9 9.5
Which application is in trouble?
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 25
Example: 100 Numbers
Start with what you haveYour measurement tool produced 100 samples
The samples areSingle applicationUser-level response time measurementsOne hour period of observation
Is the application operating well?18.1022.5021.3319.69
5.202.652.241.80
1.873.463.343.28
14.831.266.0215.00
1.777.572.205.85
3.857.986.382.38
2.772.491.367.50
4.115.331.742.76
1.464.531.7630.54
30.082.952.703.74
16.373.832.326.09
3.223.691.992.02
3.424.644.903.87
2.4815.352.131.64
2.352.203.675.60
10.462.151.461.61
23.566.216.397.38
4.294.903.542.78
3.993.283.787.37
2.943.738.8314.55
11.9813.204.752.99
3.849.768.4412.56
4.752.224.5016.67
2.513.3313.2559.55
54.503.3616.896.45
1
©2007, Apdex Alliance, Inc. All rights reserved. Slide 26
Numbers Beget Numbers
0
5
10
15
20
25
30
35
40
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60
Incremental Time Period (sec)
Num
ber o
f Sam
ples
in T
ime
Perio
d
secAverage 7.6Median 3.9Mode 4.8Standard Deviation 9.495th Percentile 22.5Minimum 1.3Maximum 59.6
Now you have 137 numbers.Can you answer the question,“Is the application operating well?”
2
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 27
Apdex Defined
Apdex is a numerical measure of user satisfaction with the performance of enterprise applicationsIt defines a method that converts many measurements into one number
Uniform 0-1 scale, 0 = no users satisfied, 1 = all users satisfied
Standardized methodIt is a comparable metric across all applications, andAcross enterprises
©2007, Apdex Alliance, Inc. All rights reserved. Slide 28
Deconstructing Application Transactions
Session = Period of time that a user is “connected” to an application
Start theapplication
End or suspendthe application
Task = Each interaction with the application during the session
Type orchoose
Enteror click
Systemresponds
Userwaits
User readsor thinks Type Wait Read
Enteror click
Systemresponds
Process = A group of user interactions that accomplish a goal
Get new email, add an employee, check on inventory status, etc. Idle
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 29
Deconstructing Application Transactions (con’t)
Turn = Each application client-and-server software interaction needed to generate a system response
Protocol = Each TCP Open, ACK, retransmission, etc, required to operate a Turn and move Payload
Packet = Each packet as seen on the wire in support of the above
Wait
©2007, Apdex Alliance, Inc. All rights reserved. Slide 30
The Task Defined
Task response time is the elapsed time required for an application system to respond to a human user input such that the user can effectively proceed with the process they are trying to accomplish
Time when the user is waiting in order to proceedUser feels the responsiveness of the applicationLong Task time makes the user less productive
The Task is what a user cantime with a stopwatch
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 31
How Users View Application Task Performance
SatisfiedUser maintains concentrationPerformance is not a factor in the user experienceTime limit threshold is unknowingly set by users and is consistent
ToleratingConcentration is impairedPerformance is now a factor in the user experienceUser will notice how long it is taking
FrustratedPerformance is typically called unacceptableCasual user may abandon the processProduction user is very likely to stop working
©2007, Apdex Alliance, Inc. All rights reserved. Slide 32
0%
2%
4%
6%
8%
10%
12%
14%
16%
0 20 40 60 80 100 120Load Time of a Typical Business Page (sec)
Prob
abili
ty o
f Exp
erie
ncin
g th
e Ti
me
Example
52% Satisfied
42% Tolerating
6% Frustrated
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 33
How Apdex Works
Start with a sufficient number of Task measurement samplesTarget response time “T” defines the satisfied zone (0-T sec)
T is shown as a subscript of all Apdex values (for example 0.80T )
Count the number of samples within three performance zonesSatisfied, Tolerating, Frustrated
Tolerating count
ApdexT = Total samples
Satisfied count 2+
GivenTarget response time T andSufficient response time measurement samples
Then
NoteFrustrated samples are not in numeratorbut are counted in total samples
Index0 = Failure; 1 = Perfection (all users satisfied)
©2007, Apdex Alliance, Inc. All rights reserved. Slide 34
Putting it All Together
Tolerating
ApdexT= Total samples
Satisfied 2+
Frustrated
Satisfied
Tolerating
Good
Fair
Poor
Una
ccep
tabl
e
0.00T
0.50T
1.00T
0.85T
0.94T
0.70T
ExcellentReport Group:ApplicationUser GroupTime Period
Existing TaskResponse TimeMeasurement
Samples
T1
2
3
45
6F
1. Define T for the applicationT = the application target time (threshold between satisfied and tolerating users).F = threshold between tolerating and frustrated users is calculated (F = 4T).
2. Define a Report Group (details available are tool dependent).3. Extract data set from existing measurements for Report Group.4. Count the number of samples in three performance zones.5. Calculate the Apdex formula.6. Display Apdex result (T is always shown as part of the result).
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 35
The Apdex View of the 100 Numbers
User productivity is impaired if the application responds in more than 8 seconds
T = 8 sec
Apdex for the 100 measurements = 0.878
The application barely providing “Good”performance
100 numbers = 0.878
4
Good
Fair
Poor
Una
ccep
tabl
e
0.00T
0.50T
1.00T
0.85T
0.94T
0.70T
Excellent
©2007, Apdex Alliance, Inc. All rights reserved. Slide 36
0%
2%
4%
6%
8%
10%
12%
14%
16%
0 20 40 60 80 100 120Load Time of a Typical Business Page (sec)
Prob
abili
ty o
f Exp
erie
ncin
g th
e Ti
me
Apdex Highlights the Long-Tail
52% Satisfied
42% Tolerating
6% Frustrated
Major eCommerce site ($4B annual on-line sales)North American broadband users accessing the San Francisco data center
This site had an average Keynote responsetime of 4 seconds, so it looked like all was wellBut: Apdex = 0.7310 = Fair
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 37
Apdex Benefits
ProcessForces a process by which the enterprise becomes focused on the important performance management issues
SimplicityConverts mountains of existing response time measurements into asimple value that can be easily understood by non-technical managers
Business LinkageOffers a clear picture of how well the IT infrastructure is really performing in support of specific business objectives
Open StandardProcesses and results that can be applied across industries and applications
©2007, Apdex Alliance, Inc. All rights reserved. Slide 38
Getting Started with Apdex
Start measuring and reportingPick any application and use any tool you can– 17+ vendors can supply the data necessary to calculate Apdex reports– Currently 5 vendors generate Apdex reports within their product
Getting some data is better than no dataIt serves as a foundation for planning and budgeting
Full benefit of Apdex requires a processInvolve key stakeholdersFormal dialog on Apdex termsBenchmark your APM capabilities before and after an Apdex pilotPlan for application SLAs based on ApdexSet the stage for continual APM quality improvement
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 39
Apdex Methodology
Gather
• Measure• Gather baseline data• Test Apdex Parameters
A T
1) Work withusers to set T
2) Work with businessmanagers to define theservice objective for A
Agree Improve
• Apdex reports & trends• Improve performance
where needed• Continual performance
& process improvement
©2007, Apdex Alliance, Inc. All rights reserved. Slide 40
The Apdex Alliance
Apdex Alliance is a non-profit industry allianceOpen collaborative approachwww.apdex.orgContributing Member – Annual dues
Supporting Member – Free– Individual interested in applying Apdex within their organization– Currently more than 700 members
Apdex is FREEYou can use Apdex in your organization or a commercial productYou or your organization do not need to join Apdex to use ApdexYou join to learn more and be plugged into the community
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 41
Leveraging the Apdex Alliance
Contributing MembersAre promoted as sponsors of the AllianceReceive support in implementing Apdex reporting
All Members (Contributing and Supporting)Receive a quarterly email newsletterHave access to formal documents and news on the web siteCan participate in the Apdex Exchange (Google Group) discussion board– All members can choose to join the discussion board– You can receive all postings, email summaries, or only see material
when you log in– There are more than 100 people in the Exchange
The numbers:7 Contributing, 700 Supporting, 100 in the Exchange
©2007, Apdex Alliance, Inc. All rights reserved. Slide 42
Apdex Alliance Member Survey
2007 Web-based survey70 responses
Number of employees15% less than 50037% 500 to 10,00048% more than 10,000
Manufacturing
IT Product Manufacturing
FinancialServices
Government
TransportationEducation & Research
Healthcare
Carriers& Utilities
Business & IT Services
Other
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 43
Who Should Lead the Apdex Alliance?
Vendors, 4%Enterprises, 11%
Balanced, 86%
©2007, Apdex Alliance, Inc. All rights reserved. Slide 44
Apdex In Use
No Plans, 26%
Likely in 1 Year, 29%Experimenting, 21%
Pilot Project, 9%
In Regular Use, 10%
Our StandardReport, 5%
45%UsingApdexNow
45%UsingApdexNow
Session 40A Application Performance and How Apdex Makes it Better
©2007, Apdex Alliance, Inc. All rights reserved. Slide 45
The Big Picture
ManageAsset
Efficiency
TSMTSM
APMAPMDeliverUserValue
Thank You
Articles and reports on performance measurement, analysis, and management are available for free at www.netforecast.com
Information about Apdex and joining the Apdex Alliance is at www.apdex.org