capacity planning and performance management for vcl clouds
TRANSCRIPT
![Page 1: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/1.jpg)
Budapest University of Technology and EconomicsDepartment of Measurement and Information Systems
Budapest University of Technology and EconomicsFault Tolerant Systems Research Group
Capacity Planning and Performance Management for VCL Clouds
รgnes Salรกnki, Gergล Kincses, Lรกszlรณ Gรถnczy, Imre Kocsis
![Page 2: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/2.jpg)
Motivation
![Page 3: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/3.jpg)
Motivation
Lab
![Page 4: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/4.jpg)
Motivation
Lab Privateuniversity cloud
![Page 5: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/5.jpg)
Motivation
Lab Privateuniversity cloud
Enterprise cloudPurchased CPU time
![Page 6: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/6.jpg)
Our VCL cloud
Maintained by our research group
5 semesters
o 2 courses/semester
9 hosts
~20 000 reservations
o Only 22 rejected
![Page 7: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/7.jpg)
Reservation Workflow in VCL
Request
o VM type
o Length
o Immediately or later
![Page 8: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/8.jpg)
Reservation Workflow in VCL
Request
o VM type
o Length
o Immediately or later
![Page 9: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/9.jpg)
Reservation Workflow in VCL
Request
o VM type
o Length
o Immediately or later
Load time
![Page 10: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/10.jpg)
Reservation Workflow in VCL
Request
o VM type
o Length
o Immediately or later
Hard reservation limit
Load time
![Page 11: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/11.jpg)
Reservation Workflow in VCL
Request
o VM type
o Length
o Immediately or later
Hard reservation limit
Load time
![Page 12: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/12.jpg)
Reservation Workflow in VCL
Request
o VM type
o Length
o Immediately or later
Hard reservation limit
Load time
![Page 13: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/13.jpg)
Capacity Planning in VCLCapacity Planning
Aron Imre
![Page 14: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/14.jpg)
Capacity Planning in VCLCapacity Planning
Aron Imre
![Page 15: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/15.jpg)
Capacity Planning in VCLCapacity Planning
Can I start solving my homework
now?
Aron ImreGergo
![Page 16: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/16.jpg)
Capacity Planning in VCLCapacity Planning
Can I start solving my homework
now?
Do we have spare capacity
for my research next week?
Aron Imre
Agnes
Gergo
![Page 17: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/17.jpg)
Capacity Planning in VCLCapacity Planning
Can I start solving my homework
now?
Do we have spare capacity
for my research next week?
I am responsible for a course with 250 students next September. Can we handle this workload?
Aron Imre
Agnes
Gergo
Laszlo
![Page 18: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/18.jpg)
Capacity Planning in VCLCapacity Planning
Support for hard limit estimation
Spare capacity prediction
Long-term capacity planning/scheduling
![Page 19: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/19.jpg)
The Available Dataset
Host1
Host2
VM1 VM2
VM1 VM2
reservation type, time to load, etc.
cpu usage, memory usage, etc.
cpu usage, memory usage, etc.
deadlines, #students
![Page 20: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/20.jpg)
Data Analysis Steps
Host1
Host2
VM1 VM2
VM1 VM2
Workload prediction
Resource util. pred.
![Page 21: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/21.jpg)
Workflow
Workload prediction
Resource util. pred.
Capacity planning
Host1
Host2
VM1 VM2
VM1 VM2
![Page 22: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/22.jpg)
Workflow
Workload prediction
Resource util. pred.
Capacity planning
Host1
Host2
VM1 VM2
VM1 VM2
Use case 1
Host
VM VM ?
![Page 23: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/23.jpg)
Workflow
Workload prediction
Resource util. pred.
Capacity planning
Host1
Host2
VM1 VM2
VM1 VM2
Use case 1
Use case 2
Host
VM VM ?
?VM VM VM
![Page 24: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/24.jpg)
Workflow
Workload prediction
Resource util. pred.
Capacity planning
Host1
Host2
VM1 VM2
VM1 VM2
Use case 1
Use case 2
Use case 3
Host
VM VM ?
?
?VM VM VM
![Page 25: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/25.jpg)
Workload prediction
![Page 26: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/26.jpg)
Workload prediction
Deadline
![Page 27: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/27.jpg)
Workload prediction
Deadline
![Page 28: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/28.jpg)
Workload prediction
Daily workload follows a Gaussian-like distribution
![Page 29: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/29.jpg)
Model fitting
![Page 30: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/30.jpg)
Workload prediction
Daily workload follows a Gaussian-like distribution
Exponential increase in peak numbers
maximum location between 7 PM and 11 PM
~4 hours as standard deviation
![Page 31: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/31.jpg)
Workload prediction
Daily workload follows a Gaussian-like distribution
Exponential increase in peak numbers
maximum location between 7 PM and 11 PM
~4 hours as standard deviation
![Page 32: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/32.jpg)
Workload prediction
![Page 33: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/33.jpg)
Workload prediction
Students work even in the night
![Page 34: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/34.jpg)
Workload prediction
Students work even in the night
They have lunch and dinner
![Page 35: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/35.jpg)
Workload prediction
Students work even in the night
They have lunch and dinner
They skip their lectures
![Page 36: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/36.jpg)
Workload prediction
Changes in studentsโ behavior?
![Page 37: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/37.jpg)
Workload prediction
Changes in studentsโ behavior?
![Page 38: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/38.jpg)
Resource Utilization Prediction
๐(๐ค๐๐๐๐๐๐๐)??
![Page 39: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/39.jpg)
Challenges
It is a cloud
o Statistical multiplexing
o Workload is not uniformly distributed
Meleg tartalรฉk
Mรกs vรกlasz ugyanarra a terhelรฉsre, pl. a memรณriรกnรกl
2012/2013/2 2013/2014/1 2013/2014/2 2014/2015/1
![Page 40: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/40.jpg)
Challenges
It is a cloud
o Statistical multiplexing
o Workload is not uniformly distributed
Meleg tartalรฉk
Mรกs vรกlasz ugyanarra a terhelรฉsre, pl. a memรณriรกnรกl
2012/2013/2 2013/2014/1 2013/2014/2 2014/2015/1
![Page 41: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/41.jpg)
Challenges
It is a cloud
o Statistical multiplexing
o Workload is not uniformly distributed
Meleg tartalรฉk
Mรกs vรกlasz ugyanarra a terhelรฉsre, pl. a memรณriรกnรกl
2012/2013/2 2013/2014/1 2013/2014/2 2014/2015/1
![Page 42: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/42.jpg)
Host1
Challenges
It is a cloud
Hosts show different behavior
![Page 43: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/43.jpg)
Host2Host1
Challenges
It is a cloud
Hosts show different behavior
![Page 44: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/44.jpg)
Host2Host1
Challenges
It is a cloud
Hosts show different behavior
o Warm spare
o Different user behavior
o ???
![Page 45: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/45.jpg)
Resource utilization analysis: memory
Linear model
o๐๐๐(๐๐1) + ๐๐๐(๐๐2) + โฆ + ๐๐๐(๐๐๐๐ก)
o Weighted by the workload
![Page 46: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/46.jpg)
Resource utilization analysis: memory
Linear model
o๐๐๐(๐๐1) + ๐๐๐(๐๐2) + โฆ + ๐๐๐(๐๐๐๐ก)
o Weighted by the workload
![Page 47: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/47.jpg)
Resource utilization analysis: memory
Linear model
o๐๐๐(๐๐1) + ๐๐๐(๐๐2) + โฆ + ๐๐๐(๐๐๐๐ก)
o Weighted by the workload
Very good at following drastic changes
![Page 48: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/48.jpg)
Resource utilization analysis: memory
Linear model
o๐๐๐(๐๐1) + ๐๐๐(๐๐2) + โฆ + ๐๐๐(๐๐๐๐ก)
o Weighted by the workload
Very good at following drastic changes
Within 5% by the 97% of time
![Page 49: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/49.jpg)
Resource utilization analysis: memory
Linear model
o๐๐๐(๐๐1) + ๐๐๐(๐๐2) + โฆ + ๐๐๐(๐๐๐๐ก)
o Weighted by the workload
![Page 50: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/50.jpg)
Resource utilization analysis: memory
Linear model
o๐๐๐(๐๐1) + ๐๐๐(๐๐2) + โฆ + ๐๐๐(๐๐๐๐ก)
o Weighted by the workload
![Page 51: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/51.jpg)
Resource utilization analysis: CPU
Linear model
o ๐ถ๐๐(๐๐1) + ๐ถ๐๐(๐๐2) + โฆ + ๐ถ๐๐(๐๐๐๐ก)
o Weighted by the workload CPU is much more sensitive than memory
![Page 52: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/52.jpg)
Resource utilization analysis: CPU
Linear model
o ๐ถ๐๐(๐๐1) + ๐ถ๐๐(๐๐2) + โฆ + ๐ถ๐๐(๐๐๐๐ก)
o Weighted by the workload
![Page 53: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/53.jpg)
Resource utilization analysis: CPU
Linear model
o ๐ถ๐๐(๐๐1) + ๐ถ๐๐(๐๐2) + โฆ + ๐ถ๐๐(๐๐๐๐ก)
o Weighted by the workload
![Page 54: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/54.jpg)
Resource utilization analysis: CPU
Linear model
o ๐ถ๐๐(๐๐1) + ๐ถ๐๐(๐๐2) + โฆ + ๐ถ๐๐(๐๐๐๐ก)
o Weighted by the workload
The students usethe CPU more
intensively beforethe deadline
![Page 55: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/55.jpg)
Resource utilization analysis: CPU
Linear model
o ๐ถ๐๐(๐๐1, ๐๐) + ๐ถ๐๐(๐๐2, ๐๐) + โฆ
o Weighted by the workload
![Page 56: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/56.jpg)
Resource utilization analysis: CPU
Linear model
o ๐ถ๐๐(๐๐1, ๐๐) + ๐ถ๐๐(๐๐2, ๐๐) + โฆ
o Weighted by the workload
![Page 57: Capacity Planning and Performance Management for VCL Clouds](https://reader031.vdocuments.us/reader031/viewer/2022030311/58eeb4a11a28ab1c428b45b1/html5/thumbnails/57.jpg)
Summary
Data-driven static capacity planning
o โuser behaviorโ analysis
o resource fingerprint estimation
Conclusions:
o student behavior can be modelled easily
o we were sometimes (too) strict
Dynamic capacity planning?
o Long loading time failed reservations soon
oWhen to burst out to a public cloud?