Recent Congestion Control Research at UCLA
Presenter: Cesar MarcondesPhD Candidate CS/UCLAChicago, July 24 2007IRTF/ICCRG Meeting
TCP Libra - RTT-Fairness for TCP
Authors: Gustavo Marfia*, Claudio E. Palazzi**, G. Pau*, Mario Gerla*, M. Y. Sanadidi*, Marco Roccetti**
NRL – UCLA*Universita` di Bologna**TCP Evaluation Suite
Authors: Hideyuki Shimonishi*, Tutomu Murase*, Cesar Marcondes**, M.Y. Sanadidi**, Mario Gerla**,
Padmanabhan Vasu**NEC Japan*
NRL – UCLA**
Recent Congestion Control Research at UCLA
Presenter: Cesar MarcondesPhD Candidate CS/UCLAChicago, July 24 2007IRTF/ICCRG Meeting
TCP Libra - RTT-Fairness for TCP
Authors: Gustavo Marfia*, Claudio E. Palazzi**, G. Pau*, Mario Gerla*, M. Y. Sanadidi*, Marco Roccetti**
NRL – UCLA*Universita` di Bologna**
IRTF/ICCRG MeetingChicago July 24 2007
Motivation & Previous WorkMotivation & Previous Work
RTT fairness: TCP sessions share
same bottleneck => same bandwidth
Not true with TCP NewReno, bandwidth share is RTT-biased
RTT fairness: TCP sessions share
same bottleneck => same bandwidth
Not true with TCP NewReno, bandwidth share is RTT-biased
TCP RTT-bias first recognized by Floyd et al, SIGCOMM 1991 Simple solution was
proposed, but proved unstable (Henderson et al)
Research community never lost interest in the RTT-fairness problem FAST BIC (Improving RTT-unfairness) Hybla
TCP RTT-bias first recognized by Floyd et al, SIGCOMM 1991 Simple solution was
proposed, but proved unstable (Henderson et al)
Research community never lost interest in the RTT-fairness problem FAST BIC (Improving RTT-unfairness) Hybla
IRTF/ICCRG MeetingChicago July 24 2007
Beyond RTT-FairnessBeyond RTT-Fairness
Other Challenges: Scalability:
TCP NewReno doesn’t scale to Gbps
Friendliness: Compatible TCP
NewReno performance (Co-existence)
Stability: Lack of Control
Theoretical Proofs of Congestion Control Protocols Stability
Other Challenges: Scalability:
TCP NewReno doesn’t scale to Gbps
Friendliness: Compatible TCP
NewReno performance (Co-existence)
Stability: Lack of Control
Theoretical Proofs of Congestion Control Protocols Stability
Libra Algorithm Solutions Scalability:
The congestion window initially grows proportional to the narrow link capacity
Friendliness: A parallel Libra goal
Stability: The congestion window
growth slows down as the MAX RTT is approached (avoid heavy congestion in the network)
Libra Algorithm Solutions Scalability:
The congestion window initially grows proportional to the narrow link capacity
Friendliness: A parallel Libra goal
Stability: The congestion window
growth slows down as the MAX RTT is approached (avoid heavy congestion in the network)
IRTF/ICCRG MeetingChicago July 24 2007
Derived directly from Fluid model:On a successful transmission:
On a packet loss:
Derived directly from Fluid model:On a successful transmission:
On a packet loss:
Libra Congestion Control Algorithm Libra Congestion Control Algorithm
€
w(n +1) = w(n) +1
w(n)α n
Tn2
To + Tn
€
w(n +1) = w(n) −w(n)
2
T1
(To + Tn )
MAX
n
Ceknϕϕ
α−
= 1
Fairness Control T0, controls
Thrput Variance& Convergence
------------T1 controls
RTT-fairness
Scalability Control C represents Estimated Capacity
and the Penalty Function is based on queueing delay
IRTF/ICCRG MeetingChicago July 24 2007
Linear-RTT Fairness Perfect-RTT Fairness Libra-RTT Fairness
Protocols with “perfect” RTT fairness:
But Hybla has narrower stability regionDelay based protocols (Vegas, Fast) have co-existence problems with TCP Reno
Libra's perfect fairness can be tuned by T1
Linear-RTT Fairness
Perfect-RTT Fairness
Libra-RTT Fairness
IRTF/ICCRG MeetingChicago July 24 2007
DumbBell TopologyDumbBell Topology
Based on BIC original Test Simulation Suite 4 forward + 4 backward
regular long-lived TCP Sack flows
25 TCP flows in both directions, window limited to 64 segments
Web traffic in both directions (20-50% of bandwidth)
Studied Connections: Short RTT conn: 21
ms Long RTT conn:
119ms
Based on BIC original Test Simulation Suite 4 forward + 4 backward
regular long-lived TCP Sack flows
25 TCP flows in both directions, window limited to 64 segments
Web traffic in both directions (20-50% of bandwidth)
Studied Connections: Short RTT conn: 21
ms Long RTT conn:
119ms
IRTF/ICCRG MeetingChicago July 24 2007
DumbBell Topology ResultsDumbBell Topology Results
Studied Flows Achieved Jain Index while competing with cross-traffic
TCP Libra obtained the best Jain’s Index specially on small buffers
Studied Flows Achieved Jain Index while competing with cross-traffic
TCP Libra obtained the best Jain’s Index specially on small buffers
Small Buffer Pipe Size Buffer
Jain’s Index
IRTF/ICCRG MeetingChicago July 24 2007
Parking Lot TopologyParking Lot Topology
Parking lot topology Flows 1 and 2: 180ms Flows 3 and 4: 90ms Flows 5 through 8:
30ms 2 buffer sizes:
375 and 2250 pkts
Jain’s Index
Parking lot topology Flows 1 and 2: 180ms Flows 3 and 4: 90ms Flows 5 through 8:
30ms 2 buffer sizes:
375 and 2250 pkts
Jain’s Index
IRTF/ICCRG MeetingChicago July 24 2007
Parking Topology ResultsParking Topology Results
Same Experiment #1Different ProtocolsJain Index Computed over Flows 1-4
Same Experiment #1Different ProtocolsJain Index Computed over Flows 1-8
Buffer Size = 375 pkts Buffer Size = Pipe Size = 2250 pkts
• TCP Libra obtained optimal RTT-Fairness among flows utilizing the
same number of congested queues
50% (even) SACK + 50% (odd) Other TCPSACK Only Point of View when Competing with Other Protocols
FAST shows unfriendlinessby reducing flows 2 and 4
BIC couldn’t reach good utilization Libra balancea:
RTT-Fairness and FriendlinessImproved RTT-Fairness of SACK !!
IRTF/ICCRG MeetingChicago July 24 2007
ConclusionConclusion The main contribution
of this work is to propose a stable solution to an old problem RTT-fairness (Floyd et al 1991)
A complete proof of the stability bound for a simple case may be found in recent publication at IFIP/Networking 2007
The main contribution of this work is to propose a stable solution to an old problem RTT-fairness (Floyd et al 1991)
A complete proof of the stability bound for a simple case may be found in recent publication at IFIP/Networking 2007
TCP Libra proves to have an excellent trade-off between fairness, efficiency and friendliness
http://www.tcplibra.org/
TCP Libra proves to have an excellent trade-off between fairness, efficiency and friendliness
http://www.tcplibra.org/
Recent Congestion Control Research at UCLA
Presenter: Cesar MarcondesPhD Candidate CS/UCLAChicago, July 24 2007IRTF/ICCRG Meeting
TCP Evaluation SuiteAuthors: Hideyuki Shimonishi*, Cesar Marcondes**, M.Y.
Sanadidi**, Mario Gerla**, Padmanabhan Vasu**NEC Japan*
NRL – UCLA**
IRTF/ICCRG MeetingChicago July 24 2007
MotivationMotivation TCP NewReno well-known doesn’t scale to Gbps
Many New Congestion Control Proposed (ARENO, Westwood, BIC, FAST, HTCP, STCP)
However, there is a need of a standard TCP Evaluation Suite for general use Lack of Meaningful Qualitative Comparison between
Proposals
TCP NewReno well-known doesn’t scale to Gbps Many New Congestion Control Proposed (ARENO, Westwood,
BIC, FAST, HTCP, STCP) However, there is a need of a standard TCP Evaluation
Suite for general use Lack of Meaningful Qualitative Comparison between
Proposals
How do they behave differently•Resource fairness vs throughput fairness•Efficiency and throughput•Fairness and friendliness
How do they co-exist with Reno•Can we have reasonable scenario for migration ?
IRTF/ICCRG MeetingChicago July 24 2007
TCP Evaluation SuiteTCP Evaluation Suite
Reproducible simulation experiments for different set of protocols Pre-configured
Environment to be used in Simulation
Core Network Size / Link Delay / Workload (Flow Size) / Start time Equal Configurable (Example) Parking Lot Topology 4 core routers w/ 2MB Links delays 15 ms (exponential) Short Lived Flow 1MB/1sec
(pareto/exponential) Long Lived Flow 4.7GB/2min
(fixed/exponential)
Reproducible simulation experiments for different set of protocols Pre-configured
Environment to be used in Simulation
Core Network Size / Link Delay / Workload (Flow Size) / Start time Equal Configurable (Example) Parking Lot Topology 4 core routers w/ 2MB Links delays 15 ms (exponential) Short Lived Flow 1MB/1sec
(pareto/exponential) Long Lived Flow 4.7GB/2min
(fixed/exponential)
Topology-generator Flow-generator Workload-generator
Link A-B BW DelaySrc A èDestB
Time : size…
Time : size…
Time : size…
Time, sizeTime, sizeÅEÅEÅE
Client-server, peer-to-peer, etc…
Simulation run 2
Compare
Heavy-tail, long-live, real trace data, etc…
Random, tree, parking-lot, etc…
Reno+Reno Reno+HS HS+HSReno+Reno Reno+HS HS+HS
SrcA èDestB
SrcA èDestBSrcA è
DestB
SrcA èDestB
SrcA èDestBLink A-B BW DelayLink A-B BW Delay
Simulation run 3Simulation run 1
IRTF/ICCRG MeetingChicago July 24 2007
Qualitative TCP Evaluation Comparison
Qualitative TCP Evaluation Comparison
After several simulations using different seeds we have the following results: RENO + RENO (today) RENO + HS (2008) HS + HS (2012)
After several simulations using different seeds we have the following results: RENO + RENO (today) RENO + HS (2008) HS + HS (2012)
TReno_iReno+Reno is the throughput of
Reno flow i in the first SET where all flow use Reno (RENO+RENO)
Where TReno_iHS+Reno is the throughput
of Reno flow i in the second SET in which half of the flows use high-speed protocols RENO + HS).
Where TiHS+HS and Ti
Reno+Reno are the throughput of flow i in the third SET where all flows use high-speed protocols (HS+HS) and the first SET where all flows use Reno (RENO+RENO), respectively
TReno_iReno+Reno is the throughput of
Reno flow i in the first SET where all flow use Reno (RENO+RENO)
Where TReno_iHS+Reno is the throughput
of Reno flow i in the second SET in which half of the flows use high-speed protocols RENO + HS).
Where TiHS+HS and Ti
Reno+Reno are the throughput of flow i in the third SET where all flows use high-speed protocols (HS+HS) and the first SET where all flows use Reno (RENO+RENO), respectively
IRTF/ICCRG MeetingChicago July 24 2007
Qualitative Comparison with TCP NewReno and Congestion Window
Dynamics Details
Qualitative Comparison with TCP NewReno and Congestion Window
Dynamics Details
Sorted by Link Utilization Sorted by Number of Hops Sorted by Flow RTT
Long-Lived Flow 1 Queue Fluctuation (10ms) Cumulative Packet Losses
IRTF/ICCRG MeetingChicago July 24 2007
The suite is available for download
The suite is available for download
http://netlab.cs.ucla.edu/tcpsuite/Topologygenerator
Flowgenerator
Workloadgenerator
Link A-B BW DelayLink A-B BW DelayScenario library
Tcl scripts for NS2
Log data
Plotting tools
New set of metrics allow qualitative comparison with the current state of the networkIt is also possible to investigate individual flow dynamics and queues reproducibly
Please submit YOUR scenario Library for public sharing !Questions, comments, contributions,to: [email protected],[email protected]