Transcript powerpoint poster - Virtuoso

Ashish Gupta, Marcia Zangrilli, Ananth I. Sundararaj,
Peter A. Dinda, Bruce B. Lowekamp
EECS, Northwestern University
Computer Science, College of William and Mary
Please visit http://virtuoso.cs.northwestern.edu
VTTIF
Free Network Measurement for
Adaptive Virtualized Distributed Computing
Virtuoso
1
Three Main Components
A Distributed Computing Platform composed
of Virtual Machines interconnected with Virtual
Networks
Major benefit : Automated Runtime Adaptation
to improve performance/cost effectiveness
VNET
VTTIF
WREN
Layer 2 virtual
overlay
networking
Runtime
application
topology
inference
Online passive
bw monitoring
and network
characterization
1. Observes incoming/outgoing packets
4. Reserve Resources when
possible
CURRENT WORK : Provides automatic
adaptation leveraging network measurements
6
4
User’s LAN
Benefits:
•Network transparency with VM migration
VM
•Ideal monitoring point for application monitoring
1. Identifies outgoing Maximal length trains with similar
spaced packets.
2. Online analysis to derive
latency/bandwidth information for all
host pair connections
2 .Calculates ISR ( Initial Sending Rate ) for these trains.
3. Answers network queries for any pair
of hosts
4. Increase trend indicates congestion, non increasing
trend indicates lower bound for bw.
User
VM to HOST mapping
Network Availability
7
UDP
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
WREN Analysis Thread
TCP
WREN Packet Tracer
Linux Kernel
Network
IP
8
Greedy Heuristic
Provide Overlay Topology
What defines good adaptation ?
 various metrics possible
5
Two approaches
Mapping
Provide forwarding rules
SOAP
Interface
bw measurements
Grid
Application
3. Monitors ACK return rate to determine trends in RTTs.
Adaptation Process
Application Demand
Latency : 20 to 100 ms , bw : 3 to 25 Mbps
VNET
How does it work ?
What does it do ?
3. Adapt distributed application
for better performance/cost
effectiveness
Nisten  emulate WAN environment with congestion
All local views aggregated to central proxy to give global
view of distributed application
WREN
2. Monitor resource availability
(bw/latency/CPU)
Controlled load/latency testbed
Resistant to rapid fluctuations and provides damped
network view
Major benefit :
Completely independent of
unmodified application or
operating system
1. Automatically infer
application demands
(network/CPU)
3
Infers application topology and traffic load at runtime
Virtual overlay network  creates illusion of LAN over wide area
ADAPTATION : A FOUR STEP
PROCESS
WREN Performance
2
Approach
Identifies 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
Simulated Annealing
Motivation : Search Space is very large  Huge
number of possibilities for mapping and overlay paths
Approach
1.Start with an initial solution
2.Perturb current configuration and evaluate with a
cost function
3. Continue Controlled Perturbation until a good cost
function is achieved
Perturbation function and algorithm details in paper
9
Adaptation Results
Scenario 1 : Only a particular mapping yields good performance
Key Advantage :
WREN accurately reports available bandwidth
when application traffic does not saturate the path
Scenario 2 : Large 256 host topology. 32 potential hosts, 8 Virtual Machines
Both Annealing and
Greedy perform well.
Results for Multi Constraint Cost
Function : Bandwidth and Latency
Annealing advantage :
Multi-Constraint
optimization easy
Annealing easy to adapt and finds
good mappings compared to
heuristic