SLA-aware load balancing for cloud datacenters
description
Transcript of SLA-aware load balancing for cloud datacenters
Copyright © 2010, [email protected]
SLA-aware load balancing for cloud datacenters
指導教授:王國禎 學生:黎中誠國立交通大學資訊工程系行動計算與寬頻網路實驗室
Copyright © 2010, [email protected]
Problem Definition
User
InternetLoad
Balancer
WAN/LAN
Real Server 1
Real Server 2 Real Server n
. . . . . . .
Cloud Computing Environment
Copyright © 2010, [email protected]
Tree of Load Balancing
Cloud load balancing
DistributedCentralized[1]
P2P[3][4]Client-Server[2]
Copyright © 2010, [email protected]
Related work
Copyright © 2010, [email protected]
Proposed Architecture
Global balancer
Local balancer Monitor
VM1 VM2 VMn...
Virtual Zone 1
Global balancer
Local balancer Monitor
VM1 VM2 VMn...
Global balancer
Local balancer Monitor
VM1 VM2 VMn...
Virtual Zone 3
Virtual Zone 2
User
Request
P2P
P2P P2P
Copyright © 2010, [email protected]
Related work
Copyright © 2010, [email protected]
Proposed Architecture
Communication
Prediction
Scheduler
Handler
History
Monitor
Adjustment
Global balancer
Local balancer
Real server1
Real server2
Real server3
Real servern
.
.
.
.
Request
P2P P2P
Copyright © 2010, [email protected]
Delta Learning Rule
Load balancer
x1c1 x2c2 x3c3
Server 1 Server 2 Server 3
weight 𝑖=𝑥𝑖∗ c𝑖
∑𝑗=1
𝑛
𝑥 𝑗∗ c 𝑗
Copyright © 2010, [email protected]
Load balancing method
• Capacity index
• Weight
C apacity index (𝐶𝑖)=1−MAX (𝐶𝑃𝑈 ,𝑀𝑒𝑚 , h𝐵𝑎𝑛𝑑𝑤𝑢𝑑𝑡 , 𝐼 /𝑂)
weight 𝑖=𝑥𝑖∗ c𝑖
∑𝑗=1
𝑛
𝑥 𝑗∗ c 𝑗
Copyright © 2010, [email protected]
Artificial neural network
• Supervised learning– Supervised learning is the machine learning
task of inferring a function from supervised (labeled) training data
• Unsupervised learning– Unsupervised learning also encompasses many
other techniques that seek to summarize and explain key features of the data
Copyright © 2010, [email protected]
Delta Learning Rule
• r = (0.8 . di-oi)f’(neti)• Δωi = η . r . x
(.) f(.)neti f(neti) oi
Learning signal generator╳ di
Δωi
η
xr
.
.
.
.
Δωi1
Δωijxj
x1
Copyright © 2010, [email protected]
Comparison of Load Balancing
[1] [2] [3] [4] Our design
Architecture Centralized Distributed Distributed Distributed Distributed
Information exchange
Connection Connection Connection Shared information
Connection
Connection Client-Server Client-Server P2P Read information
P2P
Dynamic scheduling
N Y Y Y Y
Monitor N N N Y Y
Consider the SLA
N N N N Y
Copyright © 2010, [email protected]
Conclusions
• We propose architecture based on distributed load balancer which is different from general centralized balancer
• Combination of system performance monitoring and neural network
• This system can avoid SLA violations
Copyright © 2010, [email protected]
References
• [1] V. Nae, A. Iosup, and R. Prodan, "A Scheduling Strategy on Load Balancing of Virtual Machine Resources in Cloud Computing Environment“, in Parallel and Distributed Systems, IEEE Transactions on , 2010, pp. 380 - 395.
• [2] R. Suselbeck, G. Schiele, and C. Becker, "Towards a Load Balancing in a Three-level Cloud Computing Network," in Network and Systems Support for Games (NetGames), 2009, pp. 1 - 2.
• [3] Shu-Ching Wang, Kuo-Qin Yan, Wen-Pin Liao, and Shun-Sheng Wang, "A Load Balancing Mechanism Based on Ant Colony and Complex Network Theory in Open Cloud Computing Federation," in IEEE ICCSIT, 2010, pp. 108 - 113.
• [4] Rajkumar Rajavel, "A Comparative Study into Distributed Load Balancing Algorithms for Cloud Computing," in IEEE INCOCCI, Erode, 2010, pp. 419 - 424.