Inferring the Topology and Traffic Load of Parallel Programs in a VM environment

1
Inferring the Topology and Traffic Load of
Parallel Programs in a VM environment

• Ashish Gupta
• Peter Dinda
• Department of Computer Science
• Northwestern University

2
Overview

• Motivation behind parallel programs in a VM
  environment
• Goal To infer the communication behavior
• Offline implementation
• Evaluating with parallel benchmarks
• Online Monitoring in a VM environment
• Conclusions

3
Virtuoso A VM based abstraction for a Grid
environment
4
(No Transcript)
5
Motivation

• A distributed computing environment based on
  Virtual Machines
• Raw machines connected to users network
• Our Focus Middleware support to hide the Grid
  complexity

6
(No Transcript)
7
(No Transcript)
8
Motivation

• A distributed computing environment based on
  Virtual Machines
• Raw machines connected to users network
• Our Focus Middleware support to hide the Grid
  complexity
• Our goal here Efficient execution of Parallel
  applications in such an environment

9
Parallel Application Behavior
Intelligent Placement and virtual networking of
parallel applications
Virtual Networks With VNET
VM Encapsulation
10
VNET

• Abstraction A set of VMs on same Layer 2
  network
• Virtual Ethernet LAN

11
Goal of this project

• Through low level packet traffic monitoring and
  analysis
• Inferring communication properties of parallel
  applications
• Topology
• Bandwidth requirements
• Other ?

12
Goal of this project
?
Low Level Traffic Monitoring
13
Approach
Design an offline framework
Evaluate with parallel benchmarks
If successful, design an online framework for VMs
14
An offline topology inference framework

• Goal
• A test-bed for traffic monitoring and evaluating
  topology inference methods

15
The offline method
Synced Parallel Traffic Monitoring
Traffic Filtering and Matrix Generation
Matrix Analysis and Topology Characterization
16
The offline method
Synced Parallel Traffic Monitoring
Traffic Filtering and Matrix Generation
Matrix Analysis and Topology Characterization
17
The offline method
Synced Parallel Traffic Monitoring
Traffic Filtering and Matrix Generation
Matrix Analysis and Topology Characterization
18
The offline method
Synced Parallel Traffic Monitoring
Traffic Filtering and Matrix Generation
Matrix Analysis and Topology Characterization
19
The offline method
Synced Parallel Traffic Monitoring
Traffic Filtering and Matrix Generation
Matrix Analysis and Topology Characterization
PVMPOV Inference
20
Synced Parallel Traffic Monitoring
Traffic Filtering and Matrix Generation
Matrix Analysis and Topology Characterization
Infer.pl
21
Parallel Benchmarks Evaluation

• Goal
• To test the practicality of low level traffic
  based inference

22
Parallel Benchmarks used

• Synthetic benchmarks Patterns
• N-dimensional mesh-neighbor
• N-dimensional toroid-neighbor
• N-dimensional hypercubes
• Tree reduction
• All-to-All
• Scheduling mechanism to generate deadlock free
  and efficient schemes

23
Application benchmarks

• NAS PVM benchmarks
• Popular benchmarks for parallel computing
• 5 benchmarks
• PVM-POV Distributed Ray Tracing
• Many others possible
• The inference not PVM specific
• Applicable to all communication .
• e.g. MPI, even non-parallel apps

24
Patterns application
3-D Toroid
3-D Hypercube
2-D Mesh
Reduction Tree
All-to-All
25
PVM NAS benchmarks
Parallel Integer Sort
26
Traffic Matrix for PVM IS benchmark
27
Traffic Matrix for PVM IS benchmark
Placement of host1 is crucial on the network
28
An Online Topology Inference Framework VTTIF

• Goal
• To automatically detect, monitor and report the
  global traffic matrix for a set of VMs running on
  a overlay network

29
Overall Design

• VNET
• Abstraction A set of VMs on same Layer 2
  network
• Virtual Ethernet LAN

30
A VNET virtual layer
VNET Layer
Physical Layer
31
Overall Design

• VNET
• Abstraction A set of VMs on same Layer 2
  network
• Extend VNET to include the required features
• Monitoring at Ethernet packet level
• The Challenge here
• Lacks manual control
• Detecting interesting parallel program
  communication ?

32
Detecting interesting phenomenon
Reactive Mechanisms
Proactive Mechanisms
Like a Burglar Alarm
Video Surveillance
33
Physical Host
VM
VNET daemon
VNET overlay network
Traffic Analyzer
Rate based Change detection
Traffic Matrix Query Agent
To other VNET daemons
VM Network Scheduling Agent
34
Traffic Matrix Aggregation

• Each VNET daemon keeps track of local traffic
  matrix
• Need to aggregate this information for a global
  view
• When the rate falls, the local daemons push the
  traffic matrix (When do you push the traffic
  matrix ?)
• Operation is associative reduction trees for
  scalability

The proxy daemon
35
Evaluation

• Used 4 Virtual Machines over VNET
• NAS IS benchmark

36
Conclusions
Possible to infer the topology with low level
traffic monitoring
37
Current Work

• Capabilities for dynamic adaptation into VNET
• Spatial Inference ? Network Adaptation for
  Improved Performance
• Prelim Results Improved performance upto 40 in
  execution time
• Looking into benefits of Dynamic Adaptation

38
For more information

• http//virtuoso.cs.northwestern.edu
• VNET is available for download
• PLAB web site
• plab.cs.northwestern.edu