Scribd
Upload
81 views22 pages

There Is No Such Thing As TCP: TCP: Congestion Control

This document discusses TCP congestion control algorithms including slow start, congestion avoidance, fast retransmit, and fast recovery. It explains key TCP concepts like the congestion window (cwnd), slow start threshold (ssthresh), and additive increase/multiplicative decrease. It recommends experimenting with real TCP implementations to better understand how these algorithms work in practice. The document also briefly discusses cheating TCP by opening multiple streams and potential peer-to-peer bandwidth trading approaches.

Uploaded by

feyn987
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as ODP, PDF, TXT or read online on Scribd
81 views22 pages

There Is No Such Thing As TCP: TCP: Congestion Control

This document discusses TCP congestion control algorithms including slow start, congestion avoidance, fast retransmit, and fast recovery. It explains key TCP concepts like the congestion window (cwnd), slow start threshold (ssthresh), and additive increase/multiplicative decrease. It recommends experimenting with real TCP implementations to better understand how these algorithms work in practice. The document also briefly discusses cheating TCP by opening multiple streams and potential peer-to-peer bandwidth trading approaches.

Uploaded by

feyn987
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as ODP, PDF, TXT or read online on Scribd
You are on page 1/ 22

There is no Such Thing as TCP:

TCP Congestion Control

Wolf Richter
Background
● RFC 793 – Original TCP RFC
● RFC 2001 – Close language to class
● RFC 5681 – More up-to-date RFC 2001
● Vint Cerf is here Friday
● http://dl.acm.org/citation.cfm?id=52356 –
Van Jacobson, Congestion Avoidance and Control
● Linux: man tcp
The Learning TCP Problem
● Slide's versions
● Book's version
● RFC versions
● Research paper versions
● Version in your head
● Then, there's the multiple real-world
implementations
Learn Exact Versions of TCP
● Tahoe
● Reno
● New Reno
● Vegas
● That's the goal here unfortunately
As always, experimenting on
your own with a real
implementation is the only
way you will learn anything
valuable.
So, we're making
you implement your own.
Problem: Avoid congestion
with no central coordination,
no knowledge from peers, and
no direct network feedback.
All you see are, essentially,
ACKs.
New Connection: Slow Start [Tahoe]
● Intuition: Don't flood, but quickly optimize
● Start really small: 1 SMSS
● Grow really fast: exponentially
● Occurs: beginning of TCP, after timeout
ssthresh
● cwnd – congestion window
● Governs data transmission (with rwnd)
● SMSS == sender maximum segment size
● On segment ACK, cwnd += SMSS
● ssthresh – slow start threshold
● Use slow start when cwnd < ssthresh
● Use congestion avoidance when cwnd >
ssthresh

Typically, ssthresh starts at 65535 bytes.


CA: Additive Increase
● On ACK: cwnd += SMSS*SMSS/cwnd
● Takes over when cwnd > ssthresh
● ssthresh = min(cwnd,rwnd) / 2 when
congestion
● If congestion is a timeout, cwnd = SMSS
CA: Multiplicative Decrease
● Appears depending on congestion control
● Most likely [Reno]: 3 Duplicate ACKs
● On a timeout, set cwnd = cwnd / 2
Fast Retransmit [Tahoe]
● Receiver sends duplicate ACKs
● Immediately on out-of-order segment
● Sender receives >= 3 duplicate ACKs
● Immediately retransmit segment
● cwnd = SMSS
● Slow start
● [Reno] Fast Recovery until non-duplicate
ACK
Fast Recovery [Reno, New Reno]
● ssthresh = cwnd / 2
● cwnd = ssthresh [+ 3*SMSS] (in RFC)
● Each time another duplicate ACK arrives,
● cwnd += SMSS
● Transmit new segment if allowed [New Reno]
● When ACK for new data arrives
● cwnd = ssthresh
● If timeout again, slow start with cwnd =
SMSS
Timeout Events [Tahoe, Reno]
Both treat these the same: drop to slow start
ssthresh = cwnd / 2
cwnd = SMSS
Experimenting on Your Own
● getsockopt() – on a TCP socket
● Transfer large amounts of data
● Check out TCP_INFO
● Returns a struct tcp_info;
/usr/include/netinet/tcp.h
struct tcp_info /* Times. */
u_int32_t tcpi_last_data_sent;

u_int8_ttcpi_state; u_int32_t tcpi_last_ack_sent; /* Not remembered,


sorry. */
u_int8_ttcpi_ca_state;
u_int32_t tcpi_last_data_recv;
u_int8_ttcpi_retransmits;
u_int32_t tcpi_last_ack_recv;
u_int8_ttcpi_probes;
u_int8_ttcpi_backoff; /* Metrics. */
u_int8_ttcpi_options; u_int32_t tcpi_pmtu;
u_int8_ttcpi_snd_wscale : 4, tcpi_rcv_wscale : 4; u_int32_t tcpi_rcv_ssthresh;
u_int32_t tcpi_rtt;
u_int32_t tcpi_rto; u_int32_t tcpi_rttvar;
u_int32_t tcpi_ato; u_int32_t tcpi_snd_ssthresh;

u_int32_t tcpi_snd_mss; u_int32_t tcpi_snd_cwnd;

u_int32_t tcpi_rcv_mss; u_int32_t tcpi_advmss;


u_int32_t tcpi_reordering;

u_int32_t tcpi_unacked;
u_int32_t tcpi_rcv_rtt;
u_int32_t tcpi_sacked;
u_int32_t tcpi_rcv_space;
u_int32_t tcpi_lost;
u_int32_t tcpi_retrans;
u_int32_t tcpi_total_retrans;
u_int32_t tcpi_fackets;
};
Cheating TCP: Foul Play
● What happens with two TCP streams, one
from each host, on a 10 Mbps link?
Cheating TCP: Foul Play
● What happens with two TCP streams, one
from each host, on a 10 Mbps link?
● Name them host A and host B. What if
host A opens 10 TCP streams? Host B
keeps only 1 TCP stream?
Cheating TCP: Foul Play
● What happens with two TCP streams, one
from each host, on a 10 Mbps link?
● Name them host A and host B. What if
host A opens 10 TCP streams? Host B
keeps only 1 TCP stream?
● Fair sharing across streams...
● No notion of logical peers
P2P Research: Bandwidth Trading
● UVA limited dorm links in dorm rooms
● We had high-speed WiFi between us
● What if we all colluded?
● Merging many TCP flows out-of-band :-)
● Fun senior thesis project
● P2P Bandwidth Trading (economics+CS)
LPTHW EC [10 Points]

Due November 19

Email Wolf
GitHub:

Git it, got it, good.

git clone git://github.com/theonewolf/15-441-Recitation-Sessions.git

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy