Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
1
Linux 2.4 Implementation of Westwood+ TCP with RateHalving : A Performance Evaluation
over the Internet
Angelo Dell'AeraAngelo Dell'AeraLuigi Alfredo GriecoLuigi Alfredo Grieco
Saverio MascoloSaverio Mascolo
Dipartimento di Elettrotecnica ed ElettronicaDipartimento di Elettrotecnica ed ElettronicaPolitecnico di BariPolitecnico di Bari
Via Orabona 4Via Orabona 4 70125 Bari, Italy70125 Bari, Italy
ICC2004 Paris Jun 22, 2004
Jun 22, 2004 2A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
OutlineOutline
➢ Background of TCP Westwood and TCP Background of TCP Westwood and TCP Westwood+ Westwood+
➢ EndtoEnd Bandwidth Estimation by EndtoEnd Bandwidth Estimation by filtering the ACK stream filtering the ACK stream
➢ Linux implementation of TCP Westwood+ Linux implementation of TCP Westwood+ ➢ Internet measurementsInternet measurements
Jun 22, 2004 3A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
TCP Westwood Congestion ControlTCP Westwood Congestion Control
Key idea: use endtoend bandwidth Key idea: use endtoend bandwidth estimation to estimation to adaptively set adaptively set cwndcwnd and and ssthreshssthresh after congestion after congestion
instead ofinstead of
standard standard ““blindblind” multiplicative window ” multiplicative window decreasedecrease
Jun 22, 2004 4A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Standard TCP (Van Jacobson)Standard TCP (Van Jacobson)
time
ssthresh
cwnd
Congestion Avoidance (CA)
Timeout
Fast recovery
Exponential increasing
Linear increasing
Slow Start (SS)
Typical cwnd dynamics following the AIMD paradigm
Jun 22, 2004 5A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
TCP WestwoodTCP Westwood
Congestion Avoidance
Slow start
cwnd
time
Timeoutssthresh
BWE*RTTmin
Adaptive Decrease cwnd = ssthresh = BWE * RTTmin
TCP Westwood Adaptive Decrease vs TCP (New) Reno blind by ½ window shrinking
Jun 22, 2004 6A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Known drawbacks of TCP RenoKnown drawbacks of TCP Reno
➢ Low throughput over wireless linksLow throughput over wireless linksbecause losses due to unreliable links because losses due to unreliable links are misinterpreted as congestionare misinterpreted as congestion
➢ Reno throughput is proportional to Reno throughput is proportional to 1/RTT1/RTT, i.e. it is not fair, i.e. it is not fair
Jun 22, 2004 7A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
EndtoEnd Bandwidth EndtoEnd Bandwidth EstimationEstimation
➢ TCP Westwood+ algorithm is based on TCP Westwood+ algorithm is based on endtoend estimation of the bandwidth endtoend estimation of the bandwidth available along the TCP connection pathavailable along the TCP connection path
➢ The estimate is obtained by filtering the The estimate is obtained by filtering the stream of returning ACK packets and it is stream of returning ACK packets and it is used to adaptively set the control used to adaptively set the control windows when network congestion is windows when network congestion is experienced experienced
Jun 22, 2004 8A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
EndtoEnd Bandwidth EndtoEnd Bandwidth EstimationEstimation
ACKs
packets
Filter
RECEIVERSENDER
Bandwidthestimate
ACKs
packets
Network
Jun 22, 2004 9A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
TCP Westwood+ TCP Westwood+ Pseudo CodePseudo Code
➢ When 3 DUPACKs are received by the senderWhen 3 DUPACKs are received by the sender ssthresh=max(2, (BWE * RTTmin) / MSS);ssthresh=max(2, (BWE * RTTmin) / MSS);
cwnd = ssthresh; cwnd = ssthresh;
➢ When coarse timeout expiresWhen coarse timeout expires ssthresh=max(2, (BWE * RTTmin) / MSS);ssthresh=max(2, (BWE * RTTmin) / MSS); cwnd = 1;cwnd = 1;
➢ When ACKs are successfully receivedWhen ACKs are successfully receivedcwnd increases as stated in RFC2581 cwnd increases as stated in RFC2581 the endtoend bandwidth estimate BWE is computed the endtoend bandwidth estimate BWE is computed
Jun 22, 2004 10A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
TCP Westwood+ TCP Westwood+
➢ TCP Westwood+ follows an TCP Westwood+ follows an Additive Increase Additive Increase Adaptive Decrease Adaptive Decrease paradigm paradigm
➢ TCP Westwood+ improves the stability of the TCP Westwood+ improves the stability of the standard TCP multiplicative decrease algorithm standard TCP multiplicative decrease algorithm
➢ The congestion window is decreased enough in The congestion window is decreased enough in presence of heavy congestion and not too presence of heavy congestion and not too much in presence of light congestion or losses much in presence of light congestion or losses not due to congestion not due to congestion
Jun 22, 2004 11A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
TCP Westwood+ TCP Westwood+
➢ The adaptive setting of the control windows The adaptive setting of the control windows increases the fair allocation of the available increases the fair allocation of the available bandwidth to different TCP flowsbandwidth to different TCP flows
➢ Setting Setting cwnd cwnd to BWE * RTTmin sustains a to BWE * RTTmin sustains a transmission rate ((BWE * RTTmin) / RTT) transmission rate ((BWE * RTTmin) / RTT) smaller than the bandwidth estimated at the time smaller than the bandwidth estimated at the time of congestion thus leaving room in the buffers for of congestion thus leaving room in the buffers for coexisting flowscoexisting flows
➢ This improves statistical multiplexing and fairnessThis improves statistical multiplexing and fairness
Jun 22, 2004 12A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Warning…Warning…
➢ ACKs reach the TCP sender compressedACKs reach the TCP sender compressed➢ Bandwidth samplesBandwidth samples
contain high frequency components that cannot contain high frequency components that cannot be filtered out by a discretetime filter due to be filtered out by a discretetime filter due to aliasingaliasing
1−−=
jj
jj tt
db
timealinterarriv 1 ACKtt jj =− −
Jun 22, 2004 13A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
ACK compression effectsACK compression effects
➢ ACK pairs give information about the ACK pairs give information about the bandwidth of the last link traversed on bandwidth of the last link traversed on the backward paththe backward path
➢ To smooth ACK compression we To smooth ACK compression we accumulate ACKs over an RTT and then accumulate ACKs over an RTT and then compute a bandwidth samplecompute a bandwidth sample
Jun 22, 2004 14A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
An antialiasing filter in packet networksAn antialiasing filter in packet networks
j
jj
db
∆=
RTT Lastj =∆
RTTlast in the edacknowledg dataalld j =
A n tia liased sam ples
Jun 22, 2004 15A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
EndtoEnd Bandwidth EndtoEnd Bandwidth EstimateEstimate
Internet
Anti ACK compression
Low passfilter
Segments sent
Returning ACKs
BWE
TCP Westwood+
Jun 22, 2004 16A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
➢ The following timeinvariant lowpass The following timeinvariant lowpass filter is used filter is used
kkk bbb ⋅−+⋅= − )1(ˆˆ1 αα
kRTTkd
kb =
EndtoEnd Bandwidth Estimate
Jun 22, 2004 17A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
ACK compression effectsACK compression effects
➢ We have found that We have found that ACK compressionACK compression has very important effects on TCP has very important effects on TCP
➢ ACK compression ACK compression must bemust be considered considered when doing simulation when doing simulation
Jun 22, 2004 18A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Topology with ACK compression effects Topology with ACK compression effects (10 Mbps)(10 Mbps)
20 TCP
R R
Sink
10 TCPSinks
10 TCPConnections
Reverse traffic:
10 TCP longlived NewReno
connections
Forward traffic:
20 TCP Westwood connections
Jun 22, 2004 19A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
The 20 Westwood+ connections estimate a besteffort available bandwidth that reasonably approaches the fair
share of 0.5 Mbps
1.0E+01
1.0E+02
1.0E+03
1.0E+04
1.0E+05
1.0E+06
1.0E+07
1.0E+08
1.0E+09
0 20 40 60 80 100
s
Ban
dwid
th e
stim
ate
(bps
)
Fair share
Jun 22, 2004 20A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Westwood overestimates up to 100 times the fair Westwood overestimates up to 100 times the fair share due to ACK compression share due to ACK compression
1.0E+01
1.0E+02
1.0E+03
1.0E+04
1.0E+05
1.0E+06
1.0E+07
1.0E+08
1.0E+09
0 20 40 60 80 100
s
Ba
nd
wid
th e
stim
ate
(b
ps)
Fair share
Jun 22, 2004 21A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Summary on bandwidth estimateSummary on bandwidth estimate
➢ TCP Westwood : one bandwidth sample TCP Westwood : one bandwidth sample computed for each ACK computed for each ACK (leads to bandwidth overestimate in presence (leads to bandwidth overestimate in presence of ACK compression)of ACK compression)
➢ TCP Westwood+ : one bandwidth sample TCP Westwood+ : one bandwidth sample computed for each RTTcomputed for each RTT
Jun 22, 2004 22A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Linux Implementation of TCP Westwood+Linux Implementation of TCP Westwood+
➢ Linux 2.4 TCP implementation supports the Linux 2.4 TCP implementation supports the RateHalving congestion control algorithmRateHalving congestion control algorithm
➢ The RateHalving congestion control algorithm The RateHalving congestion control algorithm adjusts the window by sending one segment adjusts the window by sending one segment per two acknowledgments for exactly one per two acknowledgments for exactly one round tripround trip
Jun 22, 2004 23A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Linux Implementation of TCP Westwood+Linux Implementation of TCP Westwood+
➢ RateHalving sets the window to exactly one RateHalving sets the window to exactly one half of the data which was actually held in half of the data which was actually held in the network during the congested round tripthe network during the congested round trip
➢ RateHalving has been slightly modified by RateHalving has been slightly modified by setting the lower bound for setting the lower bound for cwnd cwnd decrease to decrease to BWE * RTTminBWE * RTTmin
Jun 22, 2004 24A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Linux Implementation of TCP Westwood+Linux Implementation of TCP Westwood+
➢ The patch has been developed to be as less The patch has been developed to be as less intrusive as possible. In fact, it's possible to intrusive as possible. In fact, it's possible to enable/disable TCP Westwood+ through the enable/disable TCP Westwood+ through the sysctl sysctl net.ipv4.tcp_westwoodnet.ipv4.tcp_westwood
➢ The sysctl allows to switch between TCP New The sysctl allows to switch between TCP New Reno and TCP Westwood+Reno and TCP Westwood+
Jun 22, 2004 25A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Linux Implementation of TCP Westwood+Linux Implementation of TCP Westwood+
➢ TCP Westwood+ support was integrated in TCP Westwood+ support was integrated in the Linux kernel release 2.4.26pre1 (kernel the Linux kernel release 2.4.26pre1 (kernel 2.4) and kernel release 2.6.3rc1 (kernel 2.6)2.4) and kernel release 2.6.3rc1 (kernel 2.6)
http://www.kernel.orghttp://www.kernel.org http://buffer.antifork.org/westwood/westwood.htmlhttp://buffer.antifork.org/westwood/westwood.html
Jun 22, 2004 26A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Internet measurementsInternet measurements
➢ More than 4000 FTP from Bari, South Italy More than 4000 FTP from Bari, South Italy to:to:
• panther.cs.ucla.edupanther.cs.ucla.edu (UCLA) (UCLA)• ssignserv.signal.uu.seignserv.signal.uu.se (Uppsala) (Uppsala)• main.penguin.itmain.penguin.it (Parma) (Parma)
Jun 22, 2004 27A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Uploads to Uploads to panther.cs.ucla.edu panther.cs.ucla.edu (1)(1)
File size=3.2MB, From: rigel.poliba.it, To: panther.cs.ucla.edu, Total number of uploads = 197, Average New Reno Goodput = 16.86Kbyte/s,
Average Westwood+ Goodput = 25.21Kbyte/s
0
10
20
30
40
50
60
Wed Feb26
13:35:162003
Wed Feb26
19:38:032003
Wed Feb26
22:53:272003
Thu Feb 2701:41:26
2003
Thu Feb 2704:10:01
2003
Thu Feb 2706:11:28
2003
Thu Feb 2708:23:06
2003
Date
Goo
dput
(K
B/s
)
Westwood+New Reno
Jun 22, 2004 28A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Uploads to Uploads to panther.cs.ucla.edu panther.cs.ucla.edu (2)(2)
File size=32MB
0
10
20
30
40
50
60
70
80
Fri Feb 2118:15:56
2003
Sat Feb 2206:04:31
2003
Sat Feb 2220:26:44
2003
Sun Feb 2307:24:58
2003
Sun Feb 2319:06:51
2003
Mon Feb 2403:52:51
2003Date
Goodp
ut (K
B/s
)
Westwood+New Reno
Jun 22, 2004 29A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Uploads to Uploads to panther.cs.ucla.edu panther.cs.ucla.edu (3)(3)
File size=32MB
0
10
20
30
40
50
60
Fri Mar 1418:59:15
2003
Sat Mar 1506:09:27
2003
Sat Mar 1513:29:48
2003
Sat Mar 1522:20:43
2003
Sun Mar 1609:57:01
2003
Mon Mar 1702:47:05
2003Date
Goo
dput
(K
B/s
)
Westwood+New Reno
Jun 22, 2004 30A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Uploads to Uploads to panther.cs.ucla.edu panther.cs.ucla.edu (4)(4)
File size=32MB
0
10
20
30
40
50
60
70
Wed Mar19
14:15:252003
Wed Mar19
22:55:382003
Thu Mar20
02:47:102003
Thu Mar20
06:08:362003
Thu Mar20
10:45:342003
Thu Mar20
19:52:362003
Fri Mar21
01:07:102003
Fri Mar21
05:29:192003
Date
Go
od
put
(KB
/s)
Westwood+New Reno
Jun 22, 2004 31A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Main ReferencesMain References
➢ L. A. Grieco, S. MascoloL. A. Grieco, S. Mascolo “ “Performance Comparison of Reno, Vegas, and Performance Comparison of Reno, Vegas, and
Westwood+TCP Congestion Control”Westwood+TCP Congestion Control” ACM CCR Vol. 34 No. 2, April 2004
➢ S. Mascolo, C. Casetti, M. Gerla, S. Lee, M. SanadidiS. Mascolo, C. Casetti, M. Gerla, S. Lee, M. Sanadidi “ “TCP Westwood: bandwidth estimation for enhanced TCP Westwood: bandwidth estimation for enhanced
transport over wireless linkstransport over wireless links”” ACM Mobicom 2001ACM Mobicom 2001 and and Winet Journal 2002Winet Journal 2002
➢ L. A. Grieco and S. Mascolo,L. A. Grieco and S. Mascolo, “ “EndtoEnd Bandwidth Estimation for Congestion EndtoEnd Bandwidth Estimation for Congestion
Control in Packet Networks”Control in Packet Networks” Second International Workshop Second International Workshop QoSIP 2003QoSIP 2003
Jun 22, 2004 32A. Dell'Aera, L.A. Grieco, S. Mascolo – ICC2004,Paris
Thanks for the attentionThanks for the attention
Questions?Questions?