- GENI Wiki

Software-defined Networking Capabilities,
Needs in GENI for VMLab
(http://vmlab.oar.net)
Prasad Calyam; [email protected]
Sudharsan Rajagopalan; [email protected]
Programmable Networks and GENI Session, GEC15
--October 2012--
Topics of Discussion
• VMLab SDN-GENI Projects
– Capabilities, Needs for VDCloud Experiments
– Capabilities, Needs for VDC-Sim Lab Exercises
– Capabilities, Needs for OSU-MU Science DMZs
2
VDCloud Experiment in GENI
3
Test Scenario Paths Evaluated
Internet Path Route
OpenFlow Path Route
4
Test Scenario Active Measurements
5
Rutgers – GENI Connectivity
http://groups.geni.net/geni/wiki/OFRG - details OpenFlow and Internet path devices
6
Test Scenario Application Measurements
70
Internet Path
OpenFlow Path
59.6
Normalized Bandwidth %
60
50
44
40
30
20
10
4
2.2
0
Interactive Application
Interactive Applications consume
more bandwidth and take higher
task time on Internet path
Video Playback
Video Playback consumes more
bandwidth and provides higher
video quality on OpenFlow path
7
Capabilities, Needs for VDCloud Experiments
• Path switching diversity/redundancy at edge and core
– ProtoGENI nodes connected to multiple OpenFlow switches at edge,
multiple programmable path selection options in core
• Controller framework
– Proactive (pre-population of flow tables) and Reactive (flow table
updates when new flow packets arrive) flow forwarding rules setup
– Support handling of high-throughput traffic flows for handling large-scale
VD requests and cross-traffic
• Controller should not crash when cross-traffic generation is high
– Easier management and debugging of programmable flows
• Better tools to help with tedious log analysis
8
Capabilities, Needs for VDCloud Experiments (2)
• Slice I&M, management and visualization tools
– Visibility for status of reserved OpenFlow resources - bandwidth
consumption measurements of flows, end-to-end flow tracking
– Tools for easy topology configuration, adding/deleting flows, interface
configuration, sampling control of interface/flow statistics
– Access to MAC address tables for debugging purposes
– Ability to export network state and import back into new GENI slices
9
VDC-Sim
Resource allocation of
thin-clients to data centers
‘Run Simulation’
(Offline)
‘Run Experiment’
(In GENI)
Net-utility per
experiment run
10
Classroom Lab Use Cases
• Courseware for students to get hands-on experience with computer
and network virtualization concepts
– Collaboration with Sonia Fahmy, Purdue University
• Possible VDC-Sim Exercises (“Maximize Net-Utility Score”)
30+
20
30
020
– Study how cloud dynamics under different fault occurrence and crosstraffic levels affects Net-utility for different number of VD request arrivals
• E.g., compare resource allocation schemes - Provisioning: U-RAM/F-RAM
(current practice), Placement: Least Load/Least Cost/Least Latency
– Explore server-side intelligent adaptation
• E.g., write a macro script to reduce user interaction round-trips for control
actions during network health bottleneck events
– Explore client-side intelligent adaptation
• E.g., characterize thin-clients and select thin-client encodings that delivers
best QoE for different user groups – knowledge worker vs. designer/artist
– Explore network-side intelligent adaptation
• E.g., implement new OpenFlow controller application logic to handle cyberattacks via path switching and VD migrations; compare with current practice
11
Capabilities, Needs for VDC-Sim Exercises
• Many are similar to VDCloud experiment capabilities, needs
• TA sets up slice with multiple data centers and thin-clients; Student
groups work with multiple OpenFlow controllers on shared
resources
• Good documentation and examples to start from
• Other challenges… we will know when we try it! 
12
OSU Science DMZ
13
OSU-MU Experiments
• Common testbed setup tasks
– Federation of Ohio State U and U of Missouri - Columbia Science DMZs
• User accounts/roles; single sign-on; authorization policies
– End-to-end (programmable) perfSONAR instrumentation & measurement
– Establishment of VLAN extensions and GENI experimentation before
Internet2 production deployment
– Experiments with optimized large data transfers with RoCE and iWARP
• Research Use Case: Soybean translational genomics and breeding
– MU “Soybean KB” (http://soykb.org) database experiments with OSU for
set up of GENI slices to dynamically change user load patterns from
remote campuses
– Service response time analysis of distributed databases, web-services for
remote user access’ scalability
14
Capabilities, Needs for OSU-MU Science DMZs
• Many are similar to VDCloud experiment capabilities, needs
• Traffic isolation between different researcher flows
– Flow forwarding rules should handle policy-directed, independent traffic
scheduling to remote sites
• Good tools, documentation, best-practices, examples – well-equip
the Performance Engineer of the Science DMZ
• Other challenges… we will know when we try it! 
15
Thank you for your attention!
16