370 likes | 599 Views
Delay-based Congestion Control Schemes for the Internet. Pu Jian Supervisor: Prof. Mounir Hamdi HKUST 13 October 2005. Congestion Control (CC) in TCP. The stability of today's Internet relies on the congestion control built into the Transmission Control Protocol (TCP).
E N D
Delay-based Congestion Control Schemes for the Internet Pu Jian Supervisor: Prof. Mounir Hamdi HKUST 13 October 2005
Congestion Control (CC) in TCP • The stability of today's Internet relies on the congestion control built into the Transmission Control Protocol (TCP). • 1974, the original TCP is introduced, but does not contain congestion control mechanism. • 1986, congestion collapse is observed. • 1988, Van Jacobson devised the congestion control algorithm for TCP. • V. Jacobson, "Congestion avoidance and control," in Proceedings of ACM Sigcomm, 1988.
Window-based CC • TCP maintains a congestion window (cwnd) governing how many bytes of data a sender may send without acknowledgment from the receiver. • adapt cwnd to the capacity of the network • AIMD: Additive Increase Multiplicative Decrease • TCP throughput can be estimatedusing cwnd cwnd: congestion window size (Bytes) RTT: Round-Trip Time
Problem in AIMD • In AIMD, cwnd increases by 1 packet per RTT if no congestion happen • take a long time to acquire available bandwidth • e.g., 2 flows share a 10Gbps link, 200ms RTT, 1500 Bytespacket size • one flow finishes available bandwidth is 5Gbps • the other flow needs 16,667 s (or 4.63 hours) to acquire the available bandwidth
Problem inPacketLoss • Congestion-related loss: • large window requires extremely low loss rate unsustainable in practice • “For a standard TCP connection with 1500-byte packets and a 100 ms round-trip time, achieving a steady-state throughput of 10 Gbps would require … a packet drop rate of at most one congestion event every 5,000,000,000 packets (or equivalently, at most one congestion event every 1 2/3 hours).” (S. Floyd) • Non-congestion-related loss: • confuse transport protocols that assume loss as the implicit congestion signal, such as TCP • e.g., low throughput on wireless links
Problem in Packet Loss Best possible TCP with no errors (1.30 Mbps) TCP Reno (280 Kbps) Sequence number (bytes) Time (s) 2 MB TCP transfer over 2 Mbps Lucent WaveLAN (from Hari Balakrishnan)
Other Loss-based CC • TCP (AIMD): • HighSpeed TCP: • Scalable TCP (MIMD):
XCP • Explicit Control Protocol (XCP) is a congestion control protocol designed for high bandwidth-delay-product networks. • proposed by MIT people in 2002 • generalize the Explicit Congestion Notification (ECN) proposal • router-assistant • decouple utilization control from fairness control • To control congestion: use MIMD which shows fast response • To control fairness: use AIMD which converges to fairness
TCP congestion control performs poorly as bandwidth or delay increases Avg. TCP Utilization Avg. TCP Utilization 50 flows in both directions Buffer = BW x Delay RTT = 80 ms 50 flows in both directions Buffer = BW x Delay BW = 155 Mb/s Bottleneck Bandwidth (Mb/s) Round Trip Delay (sec)
Round Trip Time Round Trip Time Congestion Window Congestion Window Feedback Feedback Congestion Header How does XCP Work? Feedback = + 0.1 packet
Round Trip Time Congestion Window Feedback = + 0.1 packet How does XCP Work? Feedback = - 0.3 packet
How does XCP Work? Congestion Window = Congestion Window + Feedback XCP extends ECN Routers compute explicit feedback on window increment/decrement
Goal: Matches input traffic to link capacity & drains the queue Goal: Divides between flows to converge to fairness Looks at a flow’s state in Congestion Header Looks at aggregate traffic & queue AIMD MIMD • Algorithm: • Aggregate traffic changes by ~ Spare Bandwidth • ~ - Queue Size So, = davg Spare - Queue Algorithm: If > 0 Divide equally between flows If < 0 Divide between flows proportionally to their current rates How Does an XCP Router Compute the Feedback? Congestion Controller Fairness Controller
Algorithm: If > 0 Divide equally between flows If < 0 Divide between flows proportionally to their current rates = davg Spare - Queue Theorem:System converges to optimal utilization (i.e., stable) for any link bandwidth, delay, number of sources if: Need to estimate number of flows N RTTpkt : Round Trip Time in header Cwndpkt : Congestion Window in header T: Counting Interval (Proof based on Nyquist Criterion) the details … Congestion Controller Fairness Controller No Per-Flow State No Parameter Tuning
Performance of XCP • Simulations show that XCP outperforms TCP in both conventional and high bandwidth-delay-product environments. • fair bandwidth allocation • high utilization • small standing queue size • near-zero packet drops
and chosen to make XCP robust to delay XCP increases proportionally to spare bandwidth XCP Remains Efficient as Bandwidth or Delay Increases Utilization as a function of Bandwidth Utilization as a function of Delay Avg. Utilization Avg. Utilization Bottleneck Bandwidth (Mb/s) Round Trip Delay (sec)
Start 40 Flows Start 40 Flows Stop the 40 Flows Stop the 40 Flows XCP Shows Faster Response than TCP
(RTT is 40 ms 330 ms ) XCP is Fairer than TCP Same RTT Different RTT Avg. Throughput Avg. Throughput Flow ID Flow ID
Problem in XCP • Unfair to new or short flows • new flows get a small bandwidth allocation • prolong the duration of short flows • Complex per-packet computation • require at least 1 division and 3 multiplications per-packet (if ignore addition and subtraction) • add heavy computing burden to routers • 5 M mul/div per second for one 10 Gbps interface (if packet size 1000 Bytes)
RCP • Rate Control Protocol (RCP) is a congestion control protocol emulatingProcessor Sharing (PS) • proposed by Stanford people in 2005 • The router assigns a single rate to all flows that pass through it. • The router does not store per-flow state, and does no per-packet calculations. • feedback is computed periodically, not per-packet
Problem in RCP • Inaccurate estimation of the number of flows: N(t) = C / R(t-d0) • It is true only when the link has just been filled up by flows and queue size is 0; • Otherwise, either overestimate or underestimate the flow number. • Thus, affect convergence and link utility.
QSFCP • Quick-Start Flow Control Protocol (QSFCP) • Use the same framework as XCP and RCP • Feedbacks from routers piggy-backed on ACKs • Rate Upgrade Formula:
Evaluation of QSFCP • Don’t store any per-flow state • No per-packet calculation • Quick Start • high initial sending rate • fair to new arrival flows • shorten the duration of flows • Quick Convergence • fast response to congestion • fast redistribute available bandwidth to flows • fair to long-lived flows • e.g., multimedia flows who don’t care duration but bandwidth
List of Protocols • TCP (1988) • HighSpeed TCP (2002) • Scalable TCP (2002) • XCP (2002) • RCP (2005) • QSFCP (unpublished)
Conclusions • CC in TCP is not suitable for high-speed, high-delay, or lossy networks. • Simply making network faster doesn’t help • Flow performance (durations, throughput, etc.) is also constrained by CC protocols. • New CC is needed for the future networks. • high-speed links, wireless links, satellite links • fairness, short convergence time, high utility, low drop rate, small queue size