穨control.PDF

Size: px

Start display at page:

Download "穨control.PDF"

久姜
6 years ago
Views:

1 TCP congestion control yhmiu

2 Outline Congestion control algorithms Purpose of RFC2581 Purpose of RFC2582

3 TCP SS-DR 1998 TCP Extensions RFC SACK RFC FACK 1996 Rate-Halving 1997 OldTahoe RFC Tahoe RFC Reno RFC TCP Congestion Control RFC New-Reno RFC Vegas 1995 Pseudo-Rate 1999

4 TCP congestion control (Tahoe) Initial: cwnd = 1*segsize byte threshold = 64 KB Loop: if ( ACK received in time and cwnd <= ssthresh) cwnd += 1*segsize ; else if ( ACK received in time and cwnd > ssthresh) cwnd += segsize*segsize/cwnd + segsize/8 ; else if ( packet time out) { ssthresh = cwnd/2 ; cwnd = 1*segsize ; time out ; }

5 Congestion control algorithms (RFC2001) Slow start Congestion avoidance Fast retransmit Fast recovery

6 Fast retransmit and Fast recovery Using the number of duplicate acks receiving to decide lost or out of order After Fast retransmit algorithm sends the missing segment, Fast recovery algorithm governs the transmission of data until a non-duplicate ack arrives

7 The duplicate ack receiver SHOULD send an immediate duplicate ACK when an out-of-order segment arrives

8 Example of duplicate ack to detect lost

9 TCP congestion control(reno) Initial: cwnd = 1*segsize ssthresh = 64 KB Loop: if ( ACK received in time and cwnd < = ssthresh) cwnd+= 1*segsize ; else if ( ACK received in time and cwnd> ssthresh) cwnd+= segsize * segsize / cwnd ; /* equation 2 */ else if ( packet time out) { ssthresh = cwnd / 2 ; cwnd = 1*segsize ; time out } while (3 duplicate ACK received) { ssthresh = max( cwnd / 2, 2*segsize ); cwnd = ssthresh + 3 * segsize ; if ( a duplicate ACK received ) cwnd += 1*segsize ; else if (a non-duplicate ACK received) { cwnd = ssthresh ; break ; } } byte

10 Purpose of RFC2581 Let Reno become more general version TCP Additional considerations Restarting idle connection to Slow Start The delayed ACK algorithm can be used

11 New definition or suggestion of Reno implementation by RFC2581 (1) Initial cwnd 2*segsize (bytes) Initial ssthresh congestion When cwnd=ssthresh,the sender can use either slow start or congestion avoidance During congestion avoidance,increment of cwnd is executed on incoming non-duplicate ACK

12 New definition or suggestion of Reno implementation by RFC2581 (2) cwnd given in equation 2 will fail to increase,if cwnd is vary large Let (segsize*sgesize/cwnd) = 1byte Tahoe equation 2 actually lead to diminished performance During congestion avoidance, ack byte cwnd,cwnd segsize byte cwnd,equation 2

13 New definition or suggestion of Reno implementation by RFC2581 (3) When detect segment lost ssthresh <= max ( FS/2, 2*segsize ) When detect segment lost by timeout Set cwnd = 1*segsize

14 Restarting idle connection When TCP has not received a segment for more than one retransmission timeout To avoid TCP sending a cwnd-size line-rate burst into the network after an idle period. Set IW=min(IW, cwnd) ; then go to slow start

15 The delayed ACK algorithm For at least every second segment generating an ack An ack must be generated within 500 ms of the arrival of the first unacknowledged packet MUST NOT generate more than one ACK for every incoming segment

16 Purpose of RFC2582 Reno fast recovery multiple-packet-loss

17 example Case1 (single packet dropped from a window) The ack for this packet will ack all of the packets transmitted before fast retransmit was entered Case2 (multiple packet dropped from a window) The ack for the retransmitted packet will ack some but not all all of the packets transmitted before fast retransmit Case 2 called partial acknowledgement

18 Difference form Reno and NewReno Reno under case 2 When sender receives first partial ack, it transfers form fast recovery state to congest avoidance state NewReno partial ack packet multiple-packet-loss partial ack packet fast recover

19 Initial : send_high= the initial send seqnum; While (receive 3 duplicate ack) { if (duplicate ack cover no more than send_high) break; else if (dupliacte ack cover more than send_high) { ssthresh = max(fs/2, s*segsize); recover = highest seq num transmitted; cwnd = ssthresh + 3*segsize; } } if (receive a duplicate ack) cwnd += 1*segsize; else if (receive a non-duplicate ack) { if (this ack not ack up to recover) /* partial ack */ { cwnd -= the amount of data acked; reset restransmit timer; if ( send a new segment allowed by window ) cwnd += 1*segsize; } else if (this ack ack up to recover) break; } else if (packet timeout) send_high=highest seqnum transmitted ; break; NewReno

20 Problem Why to check send_high when entering Fast retransmit state fast retransmit retransmit timeout fast retransmit sate duplicate ack send_high retransmit timer

21 Problem Why to check recover when entering Fast retransmit state multiple-packet-loss partial ack

22 Reference RFC0793 RFC1122 RFC2001 RFC2581 RFC2582

Microsoft Word final

Microsoft Word final 只寫答案而沒有解釋說明, 扣一半分數. 針對 63.07.72. 這個 IP address,( 以十進位表示, 要寫完整過程 ) (7%) a. 這一個 IP 屬於那個 Class 的網路? 以二進位說明 (%) 其所屬的 IP 網路表示法為何?(2%) 可用 IP 範圍?(2%) 共有幾個 IP 可用?(%) mask 的值為何?(%) b. 將此 IP 網路分成 7 subnets,subnet

穨control.PDF

穨control.PDF