[PPT]
-
Upload
cameroon45 -
Category
Technology
-
view
470 -
download
0
Transcript of [PPT]
![Page 1: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/1.jpg)
Adaptive Control of Virtualized Resources in Utility Computing Environments
HP Labs: Xiaoyun Zhu, Mustafa Uysal, Zhikui Wang, Sharad Singhal University of Waterloo: Kenneth Salem
Pradeep Padala, Kang G. ShinUniversity of Michigan
![Page 2: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/2.jpg)
2
A typical scenario in data centers
Customer A
Shared Data Center
Run auction site Run news site
Customer B
Requirements
Response time < 2s
Throughput > 100 rq/sec
Pay 100$
Requirements
Response time < 5s
Throughput > 50 rq/sec
Pay 50$
![Page 3: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/3.jpg)
3
Hosting applications
Data Center
E-mail serverLinux
Web serverLinux
Database serverLinux
Common idiom: One-to-one mapping of applications to nodes
![Page 4: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/4.jpg)
4
0
1
2
3
4
5
Time
Num
ber
of CPU
sProblem: Poor utilization
Wasted Resources
Ad-hoc resource allocation schemes waste resources
![Page 5: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/5.jpg)
5
Solution: Virtual data center
Consolidate
E-mail serverLinux
Web serverLinux
Database serverLinux
Virtualization
(Xen, OpenVZ, VMware)
E-mail serverLinux
Web serverLinux
Database serverLinux
Improved utilization using consolidation
![Page 6: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/6.jpg)
6
0
1
2
3
4
5
Time
Num
ber
of CPU
sProblem: Provisioning
Average
Peak
Wasted ResourcesBursty Load Bad response time
Provisioning for dynamic workloads is hard!
Solution: Adaptive controller
![Page 7: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/7.jpg)
7
Goals
• Good utilization
• Good performance
• QoS differentiation
Average CPU utilization = 80%
Average response time = 100ms
Gold vs. Silver customers
2:1 resources
![Page 8: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/8.jpg)
8
Outline
• Motivation
• Background
• Modeling
• Design
• Evaluation
• Conclusion
![Page 9: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/9.jpg)
9
How do we provision the customers ?
Virtualized Server I Virtualized Server II
VM I
VM II
VM III
VM IV
Web Server I
DB Server I
Web Server II
DB Server II
Auction Client
News Client
Customer A
Customer B
![Page 10: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/10.jpg)
10
What are we controlling ?
Xen scheduler
VM I
VM II
ControllerCPU Usage ?
Goals• Good performance• Good utilization• QoS differentiation
Goals met ? NO
Virtualized Server
MechanismPolicy
50%
50%
80%
20%
Set CPU shares
![Page 11: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/11.jpg)
11
Related work
• Existing research– Cluster management – Load balancing– Resource allocation & scheduling– QoS differentiation
• Our contribution: Adaptive resource control – Quantitative model of system behavior– Fine-grained, adaptive control
• No wastage of resources• High throughput, low response time• QoS differentiation
![Page 12: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/12.jpg)
12
How do we design an adaptive resource controller?
Model
Design
Experiment
Evaluate
Understand system variables
Input Output
Design controller
PI, PID, I controller …
Stress the controller
Goals met ?
A control theoretic approach to systems
![Page 13: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/13.jpg)
13
Outline
• Motivation
• Background
• Modeling
• Design
• Evaluation
• Conclusion
![Page 14: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/14.jpg)
14
QoS differentiation
Modeling a virtual data center
VM Shares
Workload
Virtualized Server I Virtualized Server II
Web server I
DB server I
Web server II
DB server
II
How to differentiate between two multi-tiered systems ?
VM utilization
Response time
Throughput
![Page 15: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/15.jpg)
15
0
10
20
30
40
50
60
70
80
90
30 40 50 60 70
WWW I share
Rati
o
Response time ratio Loss ratio
Modeling two multi-tiered systemsQoS metric
Linear
Response time ratio is more controllable than loss ratio
Non-Linear
![Page 16: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/16.jpg)
16
Outline
• Motivation
• Background
• Modeling
• Design
• Evaluation
• Conclusion
![Page 17: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/17.jpg)
17
Utilization controller: an example
Solution: Self-tuning integral controller
Set to 40%
Using 20%
Controller
Utilization 20/40*100 = 50%
Utilization
goal = 80%
Set to 25%
New Utilization 20/25*100 = 80%
• Problems– Utilization is variable– Delays and errors in sensing & setting– Stability concerns
VM
![Page 18: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/18.jpg)
18
• Adjusts to varying demand• Maintains goal utilization• Knobs to control aggression (Kp)• Proven stable [Wang DSOM’05]
Utilization controller
)1()()1()( kekKkuku i
System
Utilization goal
Self-tuning controller
--
Workload
Error in utilization
e(k-1)
Measuredutilization u(k-1)
CPU allocation u(k)
![Page 19: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/19.jpg)
19
Let there be controllers
Container
consumptions Problem: All controllers independent
Want 40%
Want 70%
110% Can’t fit
(Saturation)
Solution: Arbiter controller enforcing QoS differentiation
UtilControl for WS I
Virtualized Server I
UtilControl for WS II
UtilControl for DB I
Virtualized Server II
UtilControl for DB II
![Page 20: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/20.jpg)
20
Final controller
Arbiter Controller
Requested CPU shares
Desired response time ratio
Final CPU
shares
UtilControl for WS I
Virtualized Server I
UtilControl for WS II
UtilControl for DB I
Virtualized Server II
UtilControl for DB II
Container
consumptions
![Page 21: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/21.jpg)
21
Outline
• Motivation
• Background
• Modeling
• Design
• Evaluation
• Conclusion
![Page 22: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/22.jpg)
22
Evaluation
• Multi-tiered systems– 2 HP Proliant servers– Apache + MySQL– Xen 3.0 with SEDF scheduler
• Clients– RUBiS: auction client– 2 RUBiS clients: 500 … 1000 threads
• Can we maintain 70% QoS ratio ?
![Page 23: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/23.jpg)
23
Varying load - throughput
0
40
80
120
160
200
0 20 40 60 80 100
Sample point (every 10secs)
Th
rou
gh
pu
t (r
eq
s/s
ec
) Client I Client II
500 threads
1000 threads
![Page 24: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/24.jpg)
24
0
10
20
30
40
50
60
70
6 26
Sample point (every 10 secs)
% o
f C
PU
Web I consumption Web I share
SaturationWeb I share
0
10
20
30
40
50
60
70
80
90
6 26Sample point (every 10 secs)
% o
f C
PU
Web II consumption Web II share
Web II share
Saturation
0
10
20
30
40
50
60
70
80
90
0 20 40 60 80 100
Sample point (every 10 secs)
% o
f C
PU
Web I consumption Web I shareWeb II consumption Web II share
Varying load - control
Buffer to maintain good performance
Penalized to maintain QoS ratio
Saturation
![Page 25: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/25.jpg)
25
0102030405060708090
0 20 40 60 80Sample point (10 secs)
Res
pons
e tim
e ra
tio
Controller No Controller
Varying load – QoS ratio
Goal
Goal ratio of 70% maintained!
![Page 26: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/26.jpg)
26
Conclusion• Adaptive control of virtual data center
– Good application performance• High throughput• Low response time
– Good utilization• Maintain goal CPU utilization
– QoS differentiation• Maintain goal QoS ratio
• Project page: http://kabru.eecs.umich.edu/twiki/bin/view/Main/DynamicControl
• E-mail: [email protected]
• Questions ?
![Page 27: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/27.jpg)
27
Backup and old slides
![Page 28: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/28.jpg)
28
Enterprise data centers
• Large data centers – 100s/1000s of nodes– Shared infrastructure– Run critical applications– Should meet service levels
• Problems– Power costs– Management costs– Poor utilization– Unmet service levels
![Page 29: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/29.jpg)
29
Solution: Consolidate !
![Page 30: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/30.jpg)
30
Virtualized Server IIVirtualized Server I
Customer B
Hosting two multi-tiered systems
Web Server I
Web Server II
DB Server I
DB Server II
Customer A
Auction Client
News Client
Web Server I
Web Server II
DB Server I
DB Server II
![Page 31: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/31.jpg)
31
Varying load
Time
Workload II
Workload I
500 clients
1000 clients
![Page 32: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/32.jpg)
32
0
10
20
30
40
50
60
70
6 26
Sample point (every 10 secs)
% o
f C
PU
Web I consumption Web I share
SaturationWeb I share
![Page 33: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/33.jpg)
33
0
10
20
30
40
50
60
70
80
90
6 26Sample point (every 10 secs)
% o
f C
PU
Web II consumption Web II share
Web II share
Saturation
![Page 34: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/34.jpg)
34
0
50
100
150
200
250
20 30 40 50 60 70
Web share
Thro
ughput (r
eqs/sec)
Offeredload (500)
Realthroughput(500)Offeredload (1100)
Realthroughput(1100)
Modeling results - throughputDom0 effect
Saturation causes Real throughput < Offered throughput
Web share Throughput
![Page 35: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/35.jpg)
35
Arbiter controller features
• Is an integral controller
• Decides final shares based on QoS differentiation goals
• Integral gain: knobs for aggression
• Stable – gain value based on model
![Page 36: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/36.jpg)
36
Modeling a multi-tiered system
Workload
Web share
DB share
Web usage
DB usage
QoS metrics
• Stress the system in various scenarios• Observe all variables
Web server
DB server
Virtual Server
![Page 37: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/37.jpg)
37
Modeling results – response time
0
1000
2000
3000
4000
5000
6000
20 30 40 50 60 70
Web share
Re
sp
ns
e t
ime
(m
s)
500 Clients 1100 Clients
Dom0 effect
Web share Response time
![Page 38: [PPT]](https://reader036.fdocuments.net/reader036/viewer/2022062703/554fa34bb4c9057b298b4a42/html5/thumbnails/38.jpg)
38
Questions
?