TCP/UDP communication refresher

Transcription

1 TCP/UDP communication refresher WACREN Network Monitoring and Measurement Workshop Antoine Delvaux perfsonar developer

2 TCP Transmission Control Protocol (TCP) One of the original core protocols of the Internet Protocol suite (IP) > 90% of the Internet traffic Transport Layer Delivery of a stream of bytes between programs running on computers connected to a local area network, intranet or the public Internet 2

3 TCP TCP communication is Connection oriented Reliable Ordered Error-checked Some usages of TCP Web browsers: HTTP Mail servers: SMTP File transfer programs: FTP, rsync, ssh/scp 3

4 Connection Oriented A connection is established before any user data is transferred. If the connection cannot be established, the user program is notified. If the connection is ever interrupted, the user program is notified. 4

5 Reliable TCP uses a sequence number to identify each byte of data. Sequence number identifies the order of the bytes sent. Data can be reconstructed in order regardless: Fragmentation Disordering Packet loss that may occur during transmission For every payload byte transmitted, the sequence number is incremented. 5

6 TCP/IP Protocol Stack Applications (Telnet, HTTP, ) TCP UDP ICMP IP ARP Link Layer (e.g. Ethernet, ATM, ) Physical Layer

7 UDP User Datagram Protocol Transport Layer Simpler alternative to TCP for transmission that do not require TCP reliability and/or want to avoid TCP drawbacks (latency, jitter) Connectionless datagram service, exposes the unreliability of the network to the user s program Examples: Real time communication: voice, video Transaction oriented (query/response): DNS, SNMP, NTP Stateless (useful when many clients): streaming 7

8 TCP Segments The bloc of data that TCP asks IP to deliver is called a TCP segment. Each segment contains: Data Control information 8

9 Packet Terminology IP Packet Ethernet Frame Eth. hdr 14 byte 20 bytes 20 bytes TCPSegment TCP hdr IP hdr IP data TCP data Ethernet data MTU 1500 bytes MSS 4 b

10 TCP Segment Format 1 byte 1 byte 1 byte 1 byte Source Port Destination Port Sequence Number Acknowledgment Number offset reser. Control Window Checksum Urgent Pointer Options (if any) Data 10

11 Connection flow Client SYN ISN=X SYN ISN=Y ACK=X+1 ACK=Y+1 Server 11

12 Cumulative acknowledgement Cumulative acknowledgment: The receiver sends an acknowledgment when it has received all data preceding the acknowledged sequence number. Inefficient when packets are lost. TCP Packets Example: bytes are sent in 10 different TCP packets and the first packet is lost during transmission. The receiver cannot say that it received bytes to successfully. Thus the send may then have to resend all bytes. 12

13 Selective acknowledgement Selective acknowledgment (SACK) option is defined in RFC 2018 TCP Packets Acknowledge discontinuous blocks of packets received correctly The acknowledgement can specify a number of SACK blocks In the previous example: The receiver would send SACK with sequence numbers 1000 and 9999 The sender thus retransmit only the first packet, bytes 0 to

14 Control mechanisms in TCP Mechanisms to control the TCP flow: Flow control: To prevent that the sender overruns the receiver Controlled at the receiver side Congestion control: To prevent that the sender overloads the network Controlled at the sender side The two are applied together 14

15 Flow Control Limits the sender rate to guarantee reliable delivery. Controlled by the receiving side. The sender only sends what the receiver can handle: avoid flooding The receiver continually hints the sender on how much data can be received. When the receiving host buffer fills: the next ack contains a 0 in the window size this stop transfer and allow the data in the buffer to be processed. 15

16 TCP Congestion Control The sender has two additional parameters Congestion Window (cwnd): initial value is 1 MSS (Maximum Segment Size) counted as bytes Slow-start threshold Value (ssthresh): initial value is the advertised window size Congestion control works in two modes: slow start (cwnd < ssthresh) congestion avoidance (cwnd >= ssthresh) 16

17 Response to Congestion In most cases a packet loss in a network is due to a congested router rather than a transmission error. TCP then assumes there is congestion if it detects a packet loss. A TCP sender can detect lost packets via: timeout of retransmission timer receipt of a duplicate ACK And it reduces the siez of the sending window 17

18 Buffering TCP works by buffering data at sender and receiver Image source: 18

19 Optimising buffers TCP receiver and sender buffer sizes are crucial The send side should also allocate the same amount of memory After data has been sent on the network The sending side must hold it in memory until it has been ack d If the receiver if far away, acks will take a long time to arrive If the sender memory is small, it can saturate and block transmission 19

20 TCP receive buffer Amount of data that a computer can store without acknowledging the sender. It can limit throughput even if there is no packet loss in the network! TCP transmits data up to the buffer size before waiting for the ack Therefore the full bandwidth of the network may not always get used. 20

21 TCP Tuning Adjust the network congestion avoidance parameters for TCP Typically used over high-bandwidth, highlatency networks Long-haul links (Long Fat Networks - LFN) Intercontinental circuits Well-tuned networks can perform up to many times faster. Self tuning: TCP window scale option 21

22 Congestion window Congestion Window (cwnd): How many packets can be sent at one time The larger cwnd, the higher the throughput TCP has mechanisms to dynamically determine the size of the congestion window cwnd depends on the max buffer space allocated by the kernel The optimal size is related to the Bandwidth Delay Product (BDP) 22

23 BDP: Bandwidth Delay Product BDP = Bandwidth * Round Trip Time = 2 * Bandwidth * delay Number of bytes in flight to fill a path Max number of un-acknowledged packets on the wire Max number of simultaneous bits in transit between the transmitter and the receiver High performance networks have very large BDPs 23

24 BDP Example Ping returns RTT, for example: PING ( ) 56(84) bytes of data ping statistics packets transmitted, 10 received, 0% packet loss, time 9023ms rtt min/avg/max/mdev = / / /2.873 ms If the network interfaces are 10G ethernet, the TCP buffers should be: TCP buffer = (10 Gbit / 8 bits) * = MiB 24

25 TCP labs We ll explore practical aspects of TCP Checking the effect of loss and latency on TCP performance Looking the TCP window size Checking and setting buffers values Looking at the effects of tuning on TCP performance 25