Title: The Role of PCE in the Evolution of Transport Protocols
1The Role of PCE in the Evolution of Transport
Protocols
- Pfldnet 2005, Lyon, France
- M. Y. Medy Sanadidi
- http//www.cs.ucla.edu/medy
- http//www.cs.ucla.edu/NRL/hpi/tcpw/
2Recent Issues in Transport Protocols
- Large Pipes Utilization
- Steady state
- Start-up
- Impact of Wireless Links
- Last-hop wireless
- Multihop contention networks
- Fairness for asymmetric flows
- Protocols Co-Existence
- New Paradigms
- Voice/Video
- Store-and-forward at Transport layer (e.g. PEPs,
P2P/Overlays)
3Example Satellite/802.11 Networks
4Outline
- Path Characteristics Estimation (PCE)
- Prospects for Higher Efficiency
- Future of Friendly Co-Existence
- Addressing the New Paradigms
- Summary
5Path Characteristics Estimation (PCE)
- Characteristics of Interest
- Links capacity
- Path dynamic range, i.e. buffering capacity
- Cross traffic level, path-persistence,
responsiveness - Random loss
- Multihop wireless connectivity, contention, route
diversity - Participating Nodes
- Sources only
- Sources and Destinations
- Forwarding nodes (routers, base stations,
multihop wireless nodes)
6Sharing a Link
interface queue
Buffer space
bottleneck
backlog
residual bandwidth
Flow2
bandwidth
2 flows, red one is non-responsive
Flow1
fair share ?
Propagation Time
7A Hierarchy of Characteristics
- Achieved rate
- Delay/Dynamic Range
- Packet loss
Flow Behavior
- Intensity
- Path persistence
- Elasticity
Cross Traffic Load
- Links capacities
- Propagation times
- Buffer space
- Errors
Architecture
8Path Capacity Estimation
- Path Capacity capacity of narrow link
- Pathrate rely on packet pair dispersion
measurements followed by statistical processing
of results - CapProbe use dispersion measurements perform on
line filtering of results based on end-to-end
delay - TcpProbe an adaptation of CapProbe into TCP with
minimal sender side only changes
9CapProbe and TcpProbe
10Prospects for Higher Efficiency
- Steady State
- Congestion avoidance (FAST) stable at high
throughput, co-existence ??, and random loss
impact ?? - Scaling up congestion recovery (HSTCP, STCP)
higher throughput, but fairness and stability ?? - Scaling up congestion recovery (BIC) improves on
the above in fairness - Forwarder Based (XCP) superb, when we are done
with implementation issues - PCE reliance (TCP Westwood, TCP Peach) Peach
requires forwarder priority support, TCPW
requires good estimation at high speeds
11Using PCE
- Tahoe/Reno/NewReno estimate
- Packet loss via Dup Acks
- RTT average and variance
- Maintain a pipe size (or bandwidth-delay product)
estimate ssthresh - Vegas/FAST
- Achieved Rate and its relation to the Expected
Rate, or equivalently RTT and RTTmin, or Queuing
delay - HSTCP/STCP/BIC
- Use current window size (Expected Rate) in
addition to all items above in Reno
12Using PCE (2)
- TCPW estimates
- Packet loss and type of loss
- Narrow link capacity, or Path capacity
- Achieved Rate
- Dynamic Range resulting from buffering space
- (RTTmax-RTTmin)
- XCP measures at forwarders the actual
- Links capacities
- Load intensity
- RTT (obtained from sources)
13Large Pipes Measurements Results
14Acceptable Long Term Efficiency
15Some Difference in Completion Times
16Co-Existence at Gbps Speed
17Random Loss Impact
18Effect of Random Loss
19TCPW Mining ACK Streams for PCE
Bottleneck
packets
Receiver
Sender
ACKs
Internet
measure
- Rely on PCE ( e.g. capacity, achieved rate,
dynamic range) to determine an Eligible Rate
Estimate (ERE) - ERE is used to size the congestion window after a
packet loss
20TCPW BE (2001)
- BE Sampling
- With Saverio Mascolo (P. Bari) and Claudio
Casetti (P. Torino) - Packet pair
- a noisy estimate of achieved rate/capacity
- Provides throughput boost under random loss,
overestimates under congestion - Efficient but not friendly
21TCPW RE (2002)
- RE Sampling
- Packet train
- Fair estimate under congestion, underestimates
under random loss - Used in TCPW RE and inTCP Westwood (S. Mascolo)
- Friendly
22Adaptive Estimation in TCPW
- TCPW CRB ERE ? BE if random loss, else ERE? RE
- TCPW ABSE ERE ? RE ltX lt BE by continuously
adapting the bandwidth sample width to congestion
level - TCPW Astart use ERE to help short lived flows
- TCPW BBE ERE ? u C (1-u) RE,
- where u is a congestion measure taking into
account path dynamic range
23TCPW CRB (2002)
- Combined Rate and Bandwidth
- Binary adaptive
- Congestion measure Expected Rate/Achieved Rate
- Clarified Efficiency/Friendliness tradeoff
over a threshold ?
ssthresh, cwnd BE x RTTmin
Congestionmeasure
Packet Loss Detected
Ssthresh, cwnd RE x RTTmin
under a threshold ?
24TCPW ABSE (2002)
Under Congestion
Under No Congestion
- Adaptive Bandwidth Share Estimation
- Adapt the sample interval Tk according to
congestion level - Congestion measure, similar to Vegas
- Tk ranges from one interACK interval to
current RTT - Better Efficiency/Friendliness profile than CRB
25Helping Short Lived Connections
- Approaches
- Cached ssthresh
- Larger initial window
- PCE based Hoes TCPW Astart
- Negotiation Quick-Start
- No problems here for XCP!
26TCPW Astart (2003)
- Take advantage of ERE
- Adaptively and repeatedly reset ssthresh ? ERE
until sender window reaches estimated pipe size,
or encounters packet loss
- Includes multiple mini exponential increase,
and mini linear increase phases - cwnd grows slower as it approaches BDP
- Connection converges faster to its pipe size with
less buffer overflow, since it adapts to pipe
size and transient loading
27Astart First 20 Seconds Throughput
- Good scaling with capacity and propagation time
- Robust to buffer size variation
Bottleneck capacity 40 Mbps, Buffer BDP
RTT 100ms, Bottleneck 40 Mbps
28TCPW BBE (Work in Progress)
- With H. Shimonishi (NEC, Tokyo)
- Buffer and Bandwidth Estimation
- Estimates Capacity using TcpProbe (much more
accurate than BE!!) - Higher efficiency at higher random loss rates
(e.g. 5-10) - Estimates Dynamic Range (related to buffer size)
- Improves TCPW control as a function of congestion
- The result is higher efficiency and robust
friendliness even at small buffers!
29TCPW BBE Algorithms (ICC 2005)
- Dynamic Range estimate
- Dmax RTTcong loss - RTTmin
- Current Delay Distance
- D RTT RTTmin
- Eligible Rate estimate
- ERE u C (1-u) RE
- Note u0 if D and Dmax are small
30Opportunistic Friendliness of TCPW-BBE
If Reno under-performuse all the opportunity
provided without hurting co-existing Reno flows
TCP-Reno Sender
Receiver
RTT 40msec
0.001 loss
10M-1Gbps
Receiver
TCPW-BBE Sender
If Reno performsachieve similar to Reno
31The Future of Friendly Co-Existence
- Defining Friendliness
- TCP Friendliness
- Achieve throughput equal to that of TCP Reno
under some conditions (RTT, packet loss rate) - Problematic if Reno under-perform e.g. under
random losses - Opportunistic Friendliness
- If Reno performs, achieve similar to Reno
- If Reno under-perform use all the opportunity
provided without hurting co-existing Reno flows
32Evaluating a New Proposed ProtocolThe
Efficiency/Friendliness Profile
- Each point in the graph is obtained as follows
- N legacy flows gt
- legacy throughput tR1
- total utilization U1
- N/2 legacy, N/2 proposed flows gt
- legacy throughput tR2
- Total utilization U2
- Efficiency Improvement
- E U2 / U1
- Friendliness
- F tR2 / tR1
33E/F Profiles of TCPW BE, CRB and ABSE
34E/F Profile of Vegas
1.5
Vegas vs. NewReno (RED)
1.4
1.3
Utilization Ratio G (Efficiency)
N8
N16
1.2
N4
N24
N2
1.1
1
0.4
0.6
0.8
1
1.2
1.4
Throughput Ratio L (Friendliness)
Vegas uses fixed targeted queue length gt varying
friendliness depending on number of connections!
35Addressing New Paradigms
- Audio/Video Streaming
- Increasing portion of the total traffic with
distinct requirements - Multihop Wireless
- Difficult fundamental issues
- Store-and-forward at the Transport Layer
- Revisit early problems and new opportunities
36Continuous Media Transport
- Requirements
- Minimum bandwidth
- Upper bound on delay
- Lower reliability requirements than in FTP
- Adaptive streaming objectives
- Delivered quality
- Congestion control
- Support for adaptive coding
37Addressing Continuous Media Issues
- Issues with the standard protocols
- UDP no congestion or error control
- TCP AIMD behavior undesirable due to fluctuation
in rate, and consequently delay, and intolerance
to random loss - DCCP provides an excellent framework, recommends
TFRC as one possible protocol, but allows for
alternatives - TFRC is equation based, rate-equivalent to Reno,
with smoother delivery suitable for streaming - SCTP enables multiple streams with different
congestion control mechanisms, among other
features
38Streaming Over Wireless
- Under random loss, Reno and its rate-equivalent
TFRC, will both under-perform - Approaches, some with loss discrimination, have
been proposed - TFRC Wireless
- Combination of loss discrimination schemes,
- Multi-TFRC
- Multiple TFRC connections until link is congested
- VTP
- Rate estimation and loss discrimination
39Performance Comparison
Efficiency in presence of errors5 error rate,
single connection
Rate adaptation5 error rate, single
connectionwith on/off CBR cross traffic
40TCP over Multihop Wireless
- Packet losses due to
- Contention due to hidden terminals
- Varying channel quality
- Route collapse
- Buffer overflow ??
- Solution approaches
- Neighborhood RED
- Delayed ACK extension
- Sizing the TCP window for contention reduction
41Store Forward at the Transport Layer
- Overlays/P2P tunneling through TCP connections
- PEPs breaking ETE path into concatenated TCP
connections, e.g. satellites - New(?) Requirements
- Buffer management and priority schemes for better
ETE application protocol performance - TCP Receiver advertised window role
- Related item Prioritized TCP for QOS at the
Transport layer (TCP-LP, TCPW-LP)
42Summary
- Excellent progress by many approaches for scaling
efficiency with pipe size - Focus on PCE techniques is promising, e.g. TCPW
provides - Scalable efficiency
- Robustness to random loss
- Tunable opportunistic friendliness
- Streaming, multihop wireless, and forwarding at
the Transport layer to receive attention and make
good progress
43Steady State Characteristics (TCPW RE)
For small loss rate, TCPW has much larger
window than NewReno. More scalable!
44Fairness (TCPW RE)
For small loss rate, TCPW is more fair than
NewReno