tcp tahoe reno sack vega

8/12/2019 TCP Tahoe Reno Sack Vega

1/72

TCPTahoe, Reno, NewReno,

SACK, and Vegas


2/72

cwnd : congestion window

swnd : usable sending windowrwnd : advertised receivers windowssthresh : slow-start threshold


3/72

RFC793


4/72

No cwnd

On timeout: retransmitswnd = rwnd


5/72

TCP Tahoe


6/72


7/72


8/72

new ack:if (cwnd < sstresh)

cwnd += 1else

cwnd += 1/cwnd


9/72

timeout/3rd dup ack:retransmit all unacked

ssthresh = cwnd/2cwnd = 1


10/72

Improving TCP Tahoe:

Packets still gettingthrough in dup ack -- no

need to reset the clock!


11/72

TCP Reno


12/72

new ack:if (cwnd < sstresh)

cwnd += 1else

cwnd += 1/cwnd


13/72

timeout:retransmit 1st unacked

ssthresh = cwnd/2cwnd = 1


14/72

3rd dup ack:retransmit 1st unackedssthresh = cwnd/2

cwnd = cwnd/2 + 3


15/72

Fast Recovery:the pipe is still

almost full -- no

need to restart


16/72

subsequent dup ack:cwnd++

new ack:

cwnd = ssthresh


17/72

U U+W-1

Suppose U is lost (oldest unacked) and allother packets are not. At time t, cwnd isW, and packets [U, U+W-1] are in thepipe.


18/72

Between time t and t+RTT, we would haveretransmitted U and received W-1duplicate ACK.

U U+W-1


19/72

Between time t and t+RTT, the cwndbecomes W/2 + W-1. So we get to sendW/2 new packets during the time. (Sooncwnd is going to become W/2 anyway..)

U U+W-1 U+W/2+W-1


20/72

At time t+RTT, we receive ACK forpackets [U,U+W-1], set cwnd to W/2.

U U+W-1 U+W/2+W-1


21/72


22/72

Simulation

of TCPTahoe/Reno


23/72

R1S1 S2


24/72

Enteringqueue

Exitingqueue

ACKed


25/72


26/72


27/72


28/72


29/72


30/72


31/72

Improving TCP Reno:

Timeout if multiple lossesin a window


32/72

TCP NewReno


33/72

Fast recovery starts.

are the outstandingpackets at this time.

Partial ACK

Idea: stays in fast recovery until all havebeen ACKed.


34/72

Perhaps the nextpacket is lost?


35/72

3rd dup ack:retransmit 1st unacked

ssthresh = cwnd/2cwnd = cwnd/2 + 3

remember highest


36/72


complete ack:(all are acked)cwnd = ssthresh


37/72

partial ack:

retransmitcwnd = ssthresh (?)


38/72


39/72

Note : RFC2581/RFC2582 give theaccurate/gory details. Simplied

version is presented here (eg. cwnd vsFlightSize, update of cwnd upon partialACK).


40/72

TCP SACK


41/72

Coarse Feedback


42/72

Go-Back-N

vsSelective Repeat


43/72

Use TCP header

options to reportreceived segments.


44/72

SACK Blocks :

1st block - report most

recently received segments

subsequent blocks - repeatmost recent previous blocks


45/72

pipe : num of outstandingpackets in the path.

send only if pipe < cwnd


46/72

scoreboard : which

packets have beenreceived?


47/72

3rd dup ack:pipe = cwnd - 3

retransmit 1st unackedssthresh = cwnd/2

cwnd = cwnd/2 + 3


48/72


pipe--

(if send new packet, pipe++)


49/72

partial ack:retransmitcwnd = ssthreshpipe -= 2


50/72


51/72


52/72

Power of SACK:Which packet has left the network?

Where is the gap?Decouple when to send and what tosend.


53/72

TCP Vegas


54/72

So far,

packet loss assignal of congestion.


55/72

But, already over

congested whenpackets are dropped


56/72

What other signalsare there?


57/72

RTT

Load


58/72

Expected Sending Rate

E = cwnd/BaseRTT


59/72

BaseRTT : RTT whenno congestion

(take min measured RTT inpractice)


60/72

Actual Sending Rate

A = cwnd/RTT

RTT:


61/72

If (E-A) < alphacwnd++

else if (E-A) > betacwnd--


62/72

Intuition :(E-A) x BaseRTT

represents extrabuffers occupied in

the network


63/72

Picking alpha/beta

alpha: small but non-zeroto take advantage of available bandwidthimmediately. ( = 1/BaseRTT)


64/72

Picking alpha/beta

beta : beta-alpha shouldnot be too small to

prevent oscillation.( = 3/BaseRTT)


65/72

Deployment


66/72

Feb 2004


67/72

70%SACK capable


68/72

Where is TCP Vegas?


69/72


70/72

Problem 2.

Sensitive to RTTestimation.


71/72

TCP BIC/CUBICLinux 2.6.x


72/72

Compound TCPMS Windows Vista

tcp tahoe reno sack vega

Documents