Ashish Gupta, Marcia Zangrilli, Ananth I. Sundararaj, Peter A. Dinda, Bruce B. Lowekamp EECS, Northwestern University Computer Science, College of William and Mary

1
Ashish Gupta, Marcia Zangrilli, Ananth I.
Sundararaj, Peter A. Dinda, Bruce B.
LowekampEECS, Northwestern University Computer
Science, College of William and Mary Please visit
http//virtuoso.cs.northwestern.edu
Free Network Measurement for Adaptive Virtualized
Distributed Computing
1
Virtuoso
2
Approach
VTTIF
3
Three Main Components
A Distributed Computing Platform composed of
Virtual Machines interconnected with Virtual
Networks
Infers application topology and traffic load at
runtime Resistant to rapid fluctuations and
provides damped network view All local views
aggregated to central proxy to give global view
of distributed application
WREN
VTTIF
VNET
Major benefit Automated Runtime Adaptation to
improve performance/cost effectiveness
Online passive bw monitoring and network
characterization
Runtime application topology inference
Layer 2 virtual overlay networking
VNET
4
Users LAN

• Virtual overlay network ? creates illusion of LAN
  over wide area
• Benefits
• Network transparency with VM migration
• Ideal monitoring point for application monitoring

• ADAPTATION A FOUR STEP PROCESS
• Automatically infer application demands
  (network/CPU)
• Monitor resource availability (bw/latency/CPU)
• Adapt distributed application for better
  performance/cost effectiveness
• Reserve Resources when possible

Major benefit Completely independent of
unmodified application or operating system
VM
User
5
WREN
How does it work ?
SOAP Interface
What does it do ?
1. Identifies outgoing Maximal length trains with
similar spaced packets. 2 .Calculates ISR (
Initial Sending Rate ) for these trains. 3.
Monitors ACK return rate to determine trends in
RTTs. 4. Increase trend indicates congestion, non
increasing trend indicates lower bound for bw.
bw measurements
Grid Application
1. Observes incoming/outgoing packets 2. Online
analysis to derive latency/bandwidth information
for all host pair connections 3. Answers network
queries for any pair of hosts
WREN Analysis Thread
Linux Kernel
UDP
TCP
WREN Packet Tracer
Network
IP
Two approaches
8
Adaptation Process
7
CURRENT WORK Provides automatic adaptation
leveraging network measurements
VM to HOST mapping
Greedy Heuristic
Simulated Annealing
Network Availability
MappingIdentifies Hosts which have good
bandwidth connectivity and maps VMs over
them Overlay paths Uses adapted Dijkstra to find
widest paths depending on bandwidth demands of
application process pairs (sorted in decreasing
order) ? finds path which leaves maximum
residual bottleneck bandwidth
Motivation Search Space is very large ? Huge
number of possibilities for mapping and overlay
paths Approach 1.Start with an initial
solution2.Perturb current configuration and
evaluate with a cost function3. Continue
Controlled Perturbation until a good cost
function is achieved
Provide Overlay Topology
Application Demand
Provide forwarding rules
WREN Performance
6
What defines good adaptation ? ? various metrics
possible
Controlled load/latency testbed Nisten ? emulate
WAN environment with congestion Latency 20 to
100 ms , bw 3 to 25 Mbps
Current Metric Maximum residual bottleneck
bandwidth How can we map the processes and paths
such that (available bandwidth demanded
bandwidth) is maximized ? ? Maximum room for
performance improvement
Perturbation function and algorithm details in
paper
9
Adaptation Results
Scenario 2 Large 256 host topology. 32
potential hosts, 8 Virtual Machines
Scenario 1 Only a particular mapping yields
good performance
Results for Multi Constraint Cost Function
Bandwidth and Latency Annealing easy to adapt and
finds good mappings compared to heuristic
Both Annealing and Greedy perform well. Annealing
advantage Multi-Constraint optimization easy
Key Advantage WREN accurately reports
available bandwidth when application traffic
does not saturate the path