1
play

1 Why Why flow flow control? control? sender - PDF document

Dec 30, 2023 •446 likes •589 views

Lecture 7. Lecture 7. TCP mechanisms for: TCP mechanisms for: data transfer control / flow control error control congestion control Graphical examples (applet java) of several algorithms at:

  1. Lecture 7. Lecture 7. TCP mechanisms for: TCP mechanisms for: � data transfer control / flow control � error control � congestion control Graphical examples (applet java) of several algorithms at: http://www.ce.chalmers.se/~fcela/tcp-tour.html Giuseppe Bianchi Data transfer control over TCP Data transfer control over TCP a double- a double -face face issue issue: : � ������������������ � ����������� � ������������ � �������� �!"#���$$$ � ������������� �� ���� � ������������� �� ���� ������� ������� �������� ������� ���� �� �������� � ���������������� ��� � ������������������� % �������� ������ ���� ������������� �����%��� �������� � ������������� ���&�' ()������������� *�������� � ���� ������� + ,-��.��/�,-�#���������0 Giuseppe Bianchi TCP pipelining pipelining TCP � 1�������� )�������� 2������3� �������������4 W=6 � �������������������� �������������� 5 � � W ⋅ MSS � � = thr min C , � � RTT + MSS / C � 6������ �� &���������� ���57 Giuseppe Bianchi 1

  2. Why Why flow flow control? control? sender � ������� �������� ������ receiver � #��166�8�,9:�8�,-;<������ � =����������� �&�����8�<�9:�8�<)>,������ � ���� 5��&�� �� ����� ����?&�� ���� ;�@�166 � �������������������������� � A&�������������������&����@�������5����&�$ � A!�6��!��5! 9$�5�B7 Giuseppe Bianchi Window- -based based flow flow control control Window � �������� ��������������� ������ ���� ����� � E��� ����������� ������� ����(0� F����������&������ ��������� ���!6����� �����������G Receiver buffer From IP Application process read() Receiver window � 166�8�,9:�8�,-;<������ � ������� :&������������� 8�)-�9:�8�)-,;-������ � �.������������� ����&������C��������� � ���������������8�6�����������)-%D�8�;9:�8�;->D������ � ��������������������5��&����������������?&�� ���� ,�@�166 Giuseppe Bianchi Source port Destination port 32 bit Sequence number 32 bit acknowledgement number Header 6 bit U A P R S F R C S S Y I Window size length Reserved G K H T N N checksum Urgent pointer � ����������� ����������� �� ��������� ����������� �������� ����� � ������������ � )D�������������������� ���4�� � 1���&���&�����������������-�(����&���0��� DHHCH � 6�������&���� LastByteSent - LastByteAcked <= RcvWindow. � 58,-;<�������� � #������������������,-;<���������������4���$�$�������F��4����4/5%)G � ����������������������������,-;<��������&����������(����&������ ��������0 Giuseppe Bianchi 2

  3. What is is flow flow control control needed needed for for? ? What � Window flow control guarantees receiver buffer to be able to accept outstanding segments. � When receiver buffer full, just send back win=0 � in essence, flow control guarantees that transmission bit rate never exceed receiver rate � in average! � Note that instantaneous transmission rate is arbitrary… � as well as receiver rate is discretized (application reads) Giuseppe Bianchi Sliding window window Sliding W=3 Dynamic window based reduces to S=4 pure sliding window when receiver S=5 app is very fast in reading data… S=6 SEQ S=7 W=3 1 2 3 4 5 6 7 8 9 Window “sliding” forward Giuseppe Bianchi Dynamic Dynamic window window - - example example sender receiver Rec. Buffer Exchanged param: MSS=2K, 0 4K TCP CONN sender ISN=2047, WIN=4K EMPTY SETUP (carried by receiver SYN-ACK) Application does a 2K write 2K, seq=2048 0 4K 2K Ack=4096, win=2048 Application 0 4K does a 3K write 2K, seq=4096 FULL Ack=6144, win=0 Sender blocked Application does a 2K read Ack=6144, win=2048 0 4K Sender unblocks 2K may send last 1K 1K, seq=6144 Giuseppe Bianchi 3

  4. Performance: bounded Performance: bounded by by receiver buffer receiver buffer size size � Up to 1992, common operating systems had transmitter & receiver buffer defaulted at 4096 � e.g. SunOS 4.1.3 � way suboptimal over Ethernet LANs � raising buffer to 16384 = 40% throughput increase (Papadopulos & Parulkar, 1993) � e.g. Solaris 2.2 default � most socket APIs allow apps to set (increase) socket buffer sizes � But theoretical maximum remains W=65535 bytes… Giuseppe Bianchi Maximum achievable Maximum achievable throughput throughput (assuming ( assuming infinite infinite speed speed line…) line…) 1000 Throughput (Mbps) W = 65535 bytes 100 10 1 0 100 200 300 400 500 RTT (ms) Giuseppe Bianchi Window Scale Option Window Scale Option � Appears in SYN segment � operates only if both peers understand option � allows client & server to agree on a different W scale � specified in terms of bit shift (from 1 to 14) � maximum window: 65535 * 2 b � b=14 means max W = 1.073.725.440 bytes!! Giuseppe Bianchi 4

Recommend

More recommend