SlideShare a Scribd company logo
3/5/2019 1
CMG 12 December 5, 2012
Josep Ferrandiz, Alex Gilgur
3/5/2019
A classical Capacity Problem 2
3/5/2019
Imagine… 3
Shared Resource
Exclusive Resource
How many Boa
Constrictors do
we need?
X elephants
per second
R seconds
per elephant
Definitions: Loose But True
3/5/2019 4
1.Utilization – number of active units of exclusive resource
divided by total number of units
2.Level of Service – A metric that measures the user
experience.
• We will use the blocking probability as the level of service metric for
exclusive resources (e.g., worker threads).
3.Throughput – number of requests processed per unit of
time – a measure of demand.
4.Duration – total turnaround time of a request. It includes
resource holding time as well as idle time due to contention
and other effects that prevent resource use.
5.Traffic (concurrency) – average number of concurrent
requests in the system.
3/5/2019
How can we size the system?
5
How Many Units
do you need?
75%
utilization
99.9%
availability
Concurrent Sessions:
Peak Traffic
How do we size it?
3/5/2019
Utilization is Evil!!! 6
Peak Traffic:
75% of the sessions have U <
24.92 / 265.5 = 9.36%
25% of the time
U = Average / Total
1. Uncertainty: how much to pad?
2. CapEx & OpEx: 50% utilization = 2X the HW
3. It is only a proxy
What is the actual User Experience???
History
3/5/2019
How many telephone stations do you need at an exchange???
7
History
3/5/2019 8
1909 - Poisson Distribution
applies to random telephone call arrivals
1917 - Erlang Models
Einstein; Rutherford; Bohr; Erlang; Schroedinger; ….
3/5/2019 9
How many telephone stations do you need at an exchange???
So what’s the big deal?
http://www.erlang.com/calculator/erlb/
3/5/2019 10
How many parallel lines do you need ?
How is it relevant ?
3/5/2019 11
How can we use it all?
Are Utilization and
Erlang’s model related?
B = Er (Q, N)
Why is this useful ?
N = Er-1 (B, Q)
3/5/2019 12
It is better to have fewer larger systems
How does it work?
Solid Blue Line:
B = 0.001: Saturation at ~ 400 sessions
Dashed Red Line:
To maintain U = 70%, we grow the system, and
B reduces exponentially
3/5/2019 13
Validation
The model has been repeatedly validated for large and
small systems. Results are conclusively positive.
• The system was sized for 99.9% availability.
• Actual data were collected from system logs during a crisis.
• At ~ 5 sessions, we started getting bulk arrivals (non-Poisson).
For bulk arrivals, we
now have a similar model
How do we size it?
1. Find the target value of the BM for which we want to size the circuits.
2. Apply Little’s formula: Q(t) = X(t) * R(t)
3. Build regression Q(t) ~ BM(t)
4. Compute the Q at the target BM
5. Apply Erlang B model for the Q at the level of service determined by the blocking probability
Start
For all servers
For all applications
on each server
Done
It’s all just one big nested loop
How do we size it?
Step 1. Forecast the Business Metric
ForecastPro
Autobox
R
SAS ForecastServer
…
How do we size it?
Steps 2 & 3. Apply Little’s Law and Build Regression
How do we size it?
Steps 3 & 4. Build Quantile Regression;
Compute Q (BM*)
BM*
How do we size it?
Step 5. Apply Erlang’s Model to Size the Resource
3/5/2019 19
Mid-Tier DB ProtectionWeb
Success Stories
Sizing Protection-Tier Connections & SPC
3/5/2019 20
Success Stories
LB operation change
• Dozens of Servers
• 90% of soft limit
• Operation Improvement:
• 60% connection reduction
• 2.5X increase in CPU
• More servers
• Overall more
connections
The Problem:
• Switch from RR to LC
• One large system
• “Invisible” servers
• Count doesn’t matter
• Got the 60% connection reduction
The Solution: B = Er (Q, N)
When does it work?
 Assumptions:
◦ M/M/c/0:
 No buffer
 Holding Times: Exponential
 Arrivals: Poisson
3/5/2019
There is a sheep inside the box
21
Erlang B model
What If? #1
 Queuing:
◦ M/M/c/∞:
 Infinite buffer:
◦ Always room for an arrival
◦ FIFO
 Holding times: Exponential
 Arrivals: Poisson
3/5/2019
Proceed with caution, but it works 22
Erlang C model
What if #2
 Assumptions:
◦ M/M/c/0:
 No buffer
 Holding Times:
◦ Non-Exponential
 Arrivals:
◦ Bursty?
3/5/2019
It won’t work on the tails; but for bulk (bursty) arrivals we do have a model
23
Erlang model?
Bursts -> delays &
high concurrency
Conclusions
3/5/2019
There is no reason not to use it, but “Proceed with Caution”
24
• Utilization is Evil, but a useful proxy
• Erlang’s models:
• reflect multiplexing
• are corroborated by queuing theory
• have been around for 100+ years
• tools online and in excel
• can be modified for bulk (bursty) arrivals
• can be used to identify bottlenecks
• Be careful when playing in traffic
• If connections are the bottleneck,
• use LC vs. RR load balancing
B = Er (Q, N)
Acknowledgments
 A big thank-you goes to
 Erik Ostermueller whose CMG’11 paper inspired this one
 Antoine de Saint-Exupery for the Little Prince
josep_ferrandiz@yahoo.com
alexgilgur@gmail.com
3/5/2019 Thank you! 25
“It is only with the heart that one can see rightly. What is essential is invisible to the eye.”
― Antoine de Saint-Exupéry, The Little Prince

More Related Content

Similar to Erlang capacity for_connections_cmg_1907

Iwsm2014 performance measurement for cloud computing applications using iso...
Iwsm2014   performance measurement for cloud computing applications using iso...Iwsm2014   performance measurement for cloud computing applications using iso...
Iwsm2014 performance measurement for cloud computing applications using iso...
Nesma
 
Chap2 slides
Chap2 slidesChap2 slides
Chap2 slides
BaliThorat1
 
PraveenBOUT++
PraveenBOUT++PraveenBOUT++
PraveenBOUT++
Praveen Narayanan
 
Update on Benchmark 7
Update on Benchmark 7Update on Benchmark 7
Update on Benchmark 7
PFHub PFHub
 
Update on Benchmark 7
Update on Benchmark 7Update on Benchmark 7
Update on Benchmark 7
Daniel Wheeler
 
Cw13 0.01-final
Cw13 0.01-finalCw13 0.01-final
Cw13 0.01-final
asm100
 
Lifetime-Aware Scheduling and Power Control for MTC in LTE Networks
Lifetime-Aware Scheduling and Power Control for MTC in LTE NetworksLifetime-Aware Scheduling and Power Control for MTC in LTE Networks
Lifetime-Aware Scheduling and Power Control for MTC in LTE Networks
amin azari
 
Capacity planning in cellular network
Capacity planning in cellular networkCapacity planning in cellular network
Capacity planning in cellular network
Shrutika Oswal
 
Enhanced MPSM3 for applications to quantum biological simulations
Enhanced MPSM3 for applications to quantum biological simulationsEnhanced MPSM3 for applications to quantum biological simulations
Enhanced MPSM3 for applications to quantum biological simulations
Alexander Pozdneev
 
SC17 Panel: Energy Efficiency Gains From HPC Software
SC17 Panel: Energy Efficiency Gains From HPC SoftwareSC17 Panel: Energy Efficiency Gains From HPC Software
SC17 Panel: Energy Efficiency Gains From HPC Software
inside-BigData.com
 
DESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptx
DESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptxDESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptx
DESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptx
ssuser9e6d7e
 
rbm_final_paper
rbm_final_paperrbm_final_paper
rbm_final_paper
Sam Bean
 
Many-Objective Performance Enhancement in Computing Clusters
Many-Objective Performance Enhancement in Computing ClustersMany-Objective Performance Enhancement in Computing Clusters
Many-Objective Performance Enhancement in Computing Clusters
Tarik Reza Toha
 
@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More
@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More
@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More
SIMUL8 Corporation
 
ma112006id337
ma112006id337ma112006id337
ma112006id337
matsushimalab
 
Unit 3 part2
Unit 3 part2Unit 3 part2
Unit 3 part2
Karthik Vivek
 
Matopt
MatoptMatopt
Simulation of Heterogeneous Cloud Infrastructures
Simulation of Heterogeneous Cloud InfrastructuresSimulation of Heterogeneous Cloud Infrastructures
Simulation of Heterogeneous Cloud Infrastructures
CloudLightning
 
Lecture 5- Process Synchonization_revised.pdf
Lecture 5- Process Synchonization_revised.pdfLecture 5- Process Synchonization_revised.pdf
Lecture 5- Process Synchonization_revised.pdf
Amanuelmergia
 
Performance of Go on Multicore Systems
Performance of Go on Multicore SystemsPerformance of Go on Multicore Systems
Performance of Go on Multicore Systems
No J
 

Similar to Erlang capacity for_connections_cmg_1907 (20)

Iwsm2014 performance measurement for cloud computing applications using iso...
Iwsm2014   performance measurement for cloud computing applications using iso...Iwsm2014   performance measurement for cloud computing applications using iso...
Iwsm2014 performance measurement for cloud computing applications using iso...
 
Chap2 slides
Chap2 slidesChap2 slides
Chap2 slides
 
PraveenBOUT++
PraveenBOUT++PraveenBOUT++
PraveenBOUT++
 
Update on Benchmark 7
Update on Benchmark 7Update on Benchmark 7
Update on Benchmark 7
 
Update on Benchmark 7
Update on Benchmark 7Update on Benchmark 7
Update on Benchmark 7
 
Cw13 0.01-final
Cw13 0.01-finalCw13 0.01-final
Cw13 0.01-final
 
Lifetime-Aware Scheduling and Power Control for MTC in LTE Networks
Lifetime-Aware Scheduling and Power Control for MTC in LTE NetworksLifetime-Aware Scheduling and Power Control for MTC in LTE Networks
Lifetime-Aware Scheduling and Power Control for MTC in LTE Networks
 
Capacity planning in cellular network
Capacity planning in cellular networkCapacity planning in cellular network
Capacity planning in cellular network
 
Enhanced MPSM3 for applications to quantum biological simulations
Enhanced MPSM3 for applications to quantum biological simulationsEnhanced MPSM3 for applications to quantum biological simulations
Enhanced MPSM3 for applications to quantum biological simulations
 
SC17 Panel: Energy Efficiency Gains From HPC Software
SC17 Panel: Energy Efficiency Gains From HPC SoftwareSC17 Panel: Energy Efficiency Gains From HPC Software
SC17 Panel: Energy Efficiency Gains From HPC Software
 
DESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptx
DESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptxDESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptx
DESIGN OF ROBUST CELLULAR MANUFACTURING SYSTEM FOR DYNAMIC.pptx
 
rbm_final_paper
rbm_final_paperrbm_final_paper
rbm_final_paper
 
Many-Objective Performance Enhancement in Computing Clusters
Many-Objective Performance Enhancement in Computing ClustersMany-Objective Performance Enhancement in Computing Clusters
Many-Objective Performance Enhancement in Computing Clusters
 
@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More
@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More
@SIMUL8 Virtual User Group, September: Brian Harrington, Less is More
 
ma112006id337
ma112006id337ma112006id337
ma112006id337
 
Unit 3 part2
Unit 3 part2Unit 3 part2
Unit 3 part2
 
Matopt
MatoptMatopt
Matopt
 
Simulation of Heterogeneous Cloud Infrastructures
Simulation of Heterogeneous Cloud InfrastructuresSimulation of Heterogeneous Cloud Infrastructures
Simulation of Heterogeneous Cloud Infrastructures
 
Lecture 5- Process Synchonization_revised.pdf
Lecture 5- Process Synchonization_revised.pdfLecture 5- Process Synchonization_revised.pdf
Lecture 5- Process Synchonization_revised.pdf
 
Performance of Go on Multicore Systems
Performance of Go on Multicore SystemsPerformance of Go on Multicore Systems
Performance of Go on Multicore Systems
 

More from Alex Gilgur

INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...
INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...
INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...
Alex Gilgur
 
Informs2020 using machine learning to identify the factors of people's mobi...
Informs2020   using machine learning to identify the factors of people's mobi...Informs2020   using machine learning to identify the factors of people's mobi...
Informs2020 using machine learning to identify the factors of people's mobi...
Alex Gilgur
 
Informs2019 machine learning and data mining in identification of unhappy c...
Informs2019   machine learning and data mining in identification of unhappy c...Informs2019   machine learning and data mining in identification of unhappy c...
Informs2019 machine learning and data mining in identification of unhappy c...
Alex Gilgur
 
Measuring Community Resilience: a Bayesian Approach CESUN2018
Measuring Community Resilience: a Bayesian Approach CESUN2018Measuring Community Resilience: a Bayesian Approach CESUN2018
Measuring Community Resilience: a Bayesian Approach CESUN2018
Alex Gilgur
 
The Curse of P90
The Curse of P90The Curse of P90
The Curse of P90
Alex Gilgur
 
Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016 Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016
Alex Gilgur
 
Data Science and Predictive SPC
Data Science and Predictive SPCData Science and Predictive SPC
Data Science and Predictive SPC
Alex Gilgur
 
Time Series Forecasting Modeling CMG12
Time Series Forecasting Modeling CMG12Time Series Forecasting Modeling CMG12
Time Series Forecasting Modeling CMG12
Alex Gilgur
 
CMG15 Session 525
CMG15 Session 525 CMG15 Session 525
CMG15 Session 525
Alex Gilgur
 
CSP2014 Predictive SPC
CSP2014 Predictive SPCCSP2014 Predictive SPC
CSP2014 Predictive SPC
Alex Gilgur
 
Monte carlo and network cmg'14
Monte carlo and network cmg'14Monte carlo and network cmg'14
Monte carlo and network cmg'14
Alex Gilgur
 

More from Alex Gilgur (11)

INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...
INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...
INFORMS 2021 Social cohesion and emotion analysis of media during 2020 wildfi...
 
Informs2020 using machine learning to identify the factors of people's mobi...
Informs2020   using machine learning to identify the factors of people's mobi...Informs2020   using machine learning to identify the factors of people's mobi...
Informs2020 using machine learning to identify the factors of people's mobi...
 
Informs2019 machine learning and data mining in identification of unhappy c...
Informs2019   machine learning and data mining in identification of unhappy c...Informs2019   machine learning and data mining in identification of unhappy c...
Informs2019 machine learning and data mining in identification of unhappy c...
 
Measuring Community Resilience: a Bayesian Approach CESUN2018
Measuring Community Resilience: a Bayesian Approach CESUN2018Measuring Community Resilience: a Bayesian Approach CESUN2018
Measuring Community Resilience: a Bayesian Approach CESUN2018
 
The Curse of P90
The Curse of P90The Curse of P90
The Curse of P90
 
Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016 Performance OR Capacity #CMGimPACt2016
Performance OR Capacity #CMGimPACt2016
 
Data Science and Predictive SPC
Data Science and Predictive SPCData Science and Predictive SPC
Data Science and Predictive SPC
 
Time Series Forecasting Modeling CMG12
Time Series Forecasting Modeling CMG12Time Series Forecasting Modeling CMG12
Time Series Forecasting Modeling CMG12
 
CMG15 Session 525
CMG15 Session 525 CMG15 Session 525
CMG15 Session 525
 
CSP2014 Predictive SPC
CSP2014 Predictive SPCCSP2014 Predictive SPC
CSP2014 Predictive SPC
 
Monte carlo and network cmg'14
Monte carlo and network cmg'14Monte carlo and network cmg'14
Monte carlo and network cmg'14
 

Recently uploaded

Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...
Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...
Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...
IJECEIAES
 
132/33KV substation case study Presentation
132/33KV substation case study Presentation132/33KV substation case study Presentation
132/33KV substation case study Presentation
kandramariana6
 
Eric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball play
Eric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball playEric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball play
Eric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball play
enizeyimana36
 
Generative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of contentGenerative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of content
Hitesh Mohapatra
 
New techniques for characterising damage in rock slopes.pdf
New techniques for characterising damage in rock slopes.pdfNew techniques for characterising damage in rock slopes.pdf
New techniques for characterising damage in rock slopes.pdf
wisnuprabawa3
 
Embedded machine learning-based road conditions and driving behavior monitoring
Embedded machine learning-based road conditions and driving behavior monitoringEmbedded machine learning-based road conditions and driving behavior monitoring
Embedded machine learning-based road conditions and driving behavior monitoring
IJECEIAES
 
Engine Lubrication performance System.pdf
Engine Lubrication performance System.pdfEngine Lubrication performance System.pdf
Engine Lubrication performance System.pdf
mamamaam477
 
Manufacturing Process of molasses based distillery ppt.pptx
Manufacturing Process of molasses based distillery ppt.pptxManufacturing Process of molasses based distillery ppt.pptx
Manufacturing Process of molasses based distillery ppt.pptx
Madan Karki
 
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.pptUnit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
KrishnaveniKrishnara1
 
Iron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdf
Iron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdfIron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdf
Iron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdf
RadiNasr
 
Modelagem de um CSTR com reação endotermica.pdf
Modelagem de um CSTR com reação endotermica.pdfModelagem de um CSTR com reação endotermica.pdf
Modelagem de um CSTR com reação endotermica.pdf
camseq
 
Heat Resistant Concrete Presentation ppt
Heat Resistant Concrete Presentation pptHeat Resistant Concrete Presentation ppt
Heat Resistant Concrete Presentation ppt
mamunhossenbd75
 
ACEP Magazine edition 4th launched on 05.06.2024
ACEP Magazine edition 4th launched on 05.06.2024ACEP Magazine edition 4th launched on 05.06.2024
ACEP Magazine edition 4th launched on 05.06.2024
Rahul
 
Recycled Concrete Aggregate in Construction Part II
Recycled Concrete Aggregate in Construction Part IIRecycled Concrete Aggregate in Construction Part II
Recycled Concrete Aggregate in Construction Part II
Aditya Rajan Patra
 
Textile Chemical Processing and Dyeing.pdf
Textile Chemical Processing and Dyeing.pdfTextile Chemical Processing and Dyeing.pdf
Textile Chemical Processing and Dyeing.pdf
NazakatAliKhoso2
 
basic-wireline-operations-course-mahmoud-f-radwan.pdf
basic-wireline-operations-course-mahmoud-f-radwan.pdfbasic-wireline-operations-course-mahmoud-f-radwan.pdf
basic-wireline-operations-course-mahmoud-f-radwan.pdf
NidhalKahouli2
 
ML Based Model for NIDS MSc Updated Presentation.v2.pptx
ML Based Model for NIDS MSc Updated Presentation.v2.pptxML Based Model for NIDS MSc Updated Presentation.v2.pptx
ML Based Model for NIDS MSc Updated Presentation.v2.pptx
JamalHussainArman
 
2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf
2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf
2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf
Yasser Mahgoub
 
Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...
Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...
Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...
University of Maribor
 
ISPM 15 Heat Treated Wood Stamps and why your shipping must have one
ISPM 15 Heat Treated Wood Stamps and why your shipping must have oneISPM 15 Heat Treated Wood Stamps and why your shipping must have one
ISPM 15 Heat Treated Wood Stamps and why your shipping must have one
Las Vegas Warehouse
 

Recently uploaded (20)

Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...
Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...
Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...
 
132/33KV substation case study Presentation
132/33KV substation case study Presentation132/33KV substation case study Presentation
132/33KV substation case study Presentation
 
Eric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball play
Eric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball playEric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball play
Eric Nizeyimana's document 2006 from gicumbi to ttc nyamata handball play
 
Generative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of contentGenerative AI leverages algorithms to create various forms of content
Generative AI leverages algorithms to create various forms of content
 
New techniques for characterising damage in rock slopes.pdf
New techniques for characterising damage in rock slopes.pdfNew techniques for characterising damage in rock slopes.pdf
New techniques for characterising damage in rock slopes.pdf
 
Embedded machine learning-based road conditions and driving behavior monitoring
Embedded machine learning-based road conditions and driving behavior monitoringEmbedded machine learning-based road conditions and driving behavior monitoring
Embedded machine learning-based road conditions and driving behavior monitoring
 
Engine Lubrication performance System.pdf
Engine Lubrication performance System.pdfEngine Lubrication performance System.pdf
Engine Lubrication performance System.pdf
 
Manufacturing Process of molasses based distillery ppt.pptx
Manufacturing Process of molasses based distillery ppt.pptxManufacturing Process of molasses based distillery ppt.pptx
Manufacturing Process of molasses based distillery ppt.pptx
 
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.pptUnit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
Unit-III-ELECTROCHEMICAL STORAGE DEVICES.ppt
 
Iron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdf
Iron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdfIron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdf
Iron and Steel Technology Roadmap - Towards more sustainable steelmaking.pdf
 
Modelagem de um CSTR com reação endotermica.pdf
Modelagem de um CSTR com reação endotermica.pdfModelagem de um CSTR com reação endotermica.pdf
Modelagem de um CSTR com reação endotermica.pdf
 
Heat Resistant Concrete Presentation ppt
Heat Resistant Concrete Presentation pptHeat Resistant Concrete Presentation ppt
Heat Resistant Concrete Presentation ppt
 
ACEP Magazine edition 4th launched on 05.06.2024
ACEP Magazine edition 4th launched on 05.06.2024ACEP Magazine edition 4th launched on 05.06.2024
ACEP Magazine edition 4th launched on 05.06.2024
 
Recycled Concrete Aggregate in Construction Part II
Recycled Concrete Aggregate in Construction Part IIRecycled Concrete Aggregate in Construction Part II
Recycled Concrete Aggregate in Construction Part II
 
Textile Chemical Processing and Dyeing.pdf
Textile Chemical Processing and Dyeing.pdfTextile Chemical Processing and Dyeing.pdf
Textile Chemical Processing and Dyeing.pdf
 
basic-wireline-operations-course-mahmoud-f-radwan.pdf
basic-wireline-operations-course-mahmoud-f-radwan.pdfbasic-wireline-operations-course-mahmoud-f-radwan.pdf
basic-wireline-operations-course-mahmoud-f-radwan.pdf
 
ML Based Model for NIDS MSc Updated Presentation.v2.pptx
ML Based Model for NIDS MSc Updated Presentation.v2.pptxML Based Model for NIDS MSc Updated Presentation.v2.pptx
ML Based Model for NIDS MSc Updated Presentation.v2.pptx
 
2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf
2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf
2008 BUILDING CONSTRUCTION Illustrated - Ching Chapter 02 The Building.pdf
 
Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...
Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...
Presentation of IEEE Slovenia CIS (Computational Intelligence Society) Chapte...
 
ISPM 15 Heat Treated Wood Stamps and why your shipping must have one
ISPM 15 Heat Treated Wood Stamps and why your shipping must have oneISPM 15 Heat Treated Wood Stamps and why your shipping must have one
ISPM 15 Heat Treated Wood Stamps and why your shipping must have one
 

Erlang capacity for_connections_cmg_1907

  • 1. 3/5/2019 1 CMG 12 December 5, 2012 Josep Ferrandiz, Alex Gilgur
  • 3. 3/5/2019 Imagine… 3 Shared Resource Exclusive Resource How many Boa Constrictors do we need? X elephants per second R seconds per elephant
  • 4. Definitions: Loose But True 3/5/2019 4 1.Utilization – number of active units of exclusive resource divided by total number of units 2.Level of Service – A metric that measures the user experience. • We will use the blocking probability as the level of service metric for exclusive resources (e.g., worker threads). 3.Throughput – number of requests processed per unit of time – a measure of demand. 4.Duration – total turnaround time of a request. It includes resource holding time as well as idle time due to contention and other effects that prevent resource use. 5.Traffic (concurrency) – average number of concurrent requests in the system.
  • 5. 3/5/2019 How can we size the system? 5 How Many Units do you need? 75% utilization 99.9% availability Concurrent Sessions: Peak Traffic
  • 6. How do we size it? 3/5/2019 Utilization is Evil!!! 6 Peak Traffic: 75% of the sessions have U < 24.92 / 265.5 = 9.36% 25% of the time U = Average / Total 1. Uncertainty: how much to pad? 2. CapEx & OpEx: 50% utilization = 2X the HW 3. It is only a proxy What is the actual User Experience???
  • 7. History 3/5/2019 How many telephone stations do you need at an exchange??? 7
  • 8. History 3/5/2019 8 1909 - Poisson Distribution applies to random telephone call arrivals 1917 - Erlang Models Einstein; Rutherford; Bohr; Erlang; Schroedinger; ….
  • 9. 3/5/2019 9 How many telephone stations do you need at an exchange??? So what’s the big deal? http://www.erlang.com/calculator/erlb/
  • 10. 3/5/2019 10 How many parallel lines do you need ? How is it relevant ?
  • 11. 3/5/2019 11 How can we use it all? Are Utilization and Erlang’s model related? B = Er (Q, N) Why is this useful ? N = Er-1 (B, Q)
  • 12. 3/5/2019 12 It is better to have fewer larger systems How does it work? Solid Blue Line: B = 0.001: Saturation at ~ 400 sessions Dashed Red Line: To maintain U = 70%, we grow the system, and B reduces exponentially
  • 13. 3/5/2019 13 Validation The model has been repeatedly validated for large and small systems. Results are conclusively positive. • The system was sized for 99.9% availability. • Actual data were collected from system logs during a crisis. • At ~ 5 sessions, we started getting bulk arrivals (non-Poisson). For bulk arrivals, we now have a similar model
  • 14. How do we size it? 1. Find the target value of the BM for which we want to size the circuits. 2. Apply Little’s formula: Q(t) = X(t) * R(t) 3. Build regression Q(t) ~ BM(t) 4. Compute the Q at the target BM 5. Apply Erlang B model for the Q at the level of service determined by the blocking probability Start For all servers For all applications on each server Done It’s all just one big nested loop
  • 15. How do we size it? Step 1. Forecast the Business Metric ForecastPro Autobox R SAS ForecastServer …
  • 16. How do we size it? Steps 2 & 3. Apply Little’s Law and Build Regression
  • 17. How do we size it? Steps 3 & 4. Build Quantile Regression; Compute Q (BM*) BM*
  • 18. How do we size it? Step 5. Apply Erlang’s Model to Size the Resource
  • 19. 3/5/2019 19 Mid-Tier DB ProtectionWeb Success Stories Sizing Protection-Tier Connections & SPC
  • 20. 3/5/2019 20 Success Stories LB operation change • Dozens of Servers • 90% of soft limit • Operation Improvement: • 60% connection reduction • 2.5X increase in CPU • More servers • Overall more connections The Problem: • Switch from RR to LC • One large system • “Invisible” servers • Count doesn’t matter • Got the 60% connection reduction The Solution: B = Er (Q, N)
  • 21. When does it work?  Assumptions: ◦ M/M/c/0:  No buffer  Holding Times: Exponential  Arrivals: Poisson 3/5/2019 There is a sheep inside the box 21 Erlang B model
  • 22. What If? #1  Queuing: ◦ M/M/c/∞:  Infinite buffer: ◦ Always room for an arrival ◦ FIFO  Holding times: Exponential  Arrivals: Poisson 3/5/2019 Proceed with caution, but it works 22 Erlang C model
  • 23. What if #2  Assumptions: ◦ M/M/c/0:  No buffer  Holding Times: ◦ Non-Exponential  Arrivals: ◦ Bursty? 3/5/2019 It won’t work on the tails; but for bulk (bursty) arrivals we do have a model 23 Erlang model? Bursts -> delays & high concurrency
  • 24. Conclusions 3/5/2019 There is no reason not to use it, but “Proceed with Caution” 24 • Utilization is Evil, but a useful proxy • Erlang’s models: • reflect multiplexing • are corroborated by queuing theory • have been around for 100+ years • tools online and in excel • can be modified for bulk (bursty) arrivals • can be used to identify bottlenecks • Be careful when playing in traffic • If connections are the bottleneck, • use LC vs. RR load balancing B = Er (Q, N)
  • 25. Acknowledgments  A big thank-you goes to  Erik Ostermueller whose CMG’11 paper inspired this one  Antoine de Saint-Exupery for the Little Prince josep_ferrandiz@yahoo.com alexgilgur@gmail.com 3/5/2019 Thank you! 25 “It is only with the heart that one can see rightly. What is essential is invisible to the eye.” ― Antoine de Saint-Exupéry, The Little Prince