Network Research and Linux at the Hamilton Institute, NUIM.

Transcription

1 Network Research and Linux at the Hamilton Institute, NUIM. David Malone 4 November 26 1

2 What has TCP ever done for Us? Demuxes applications (using port numbers). Makes sure lost data is retransmitted. Delivers data to application in order. Engages in congestion control. Allows OOB data. Some weird stuff with TCP options. 2

3 Standard Picture of TCP Connector SYN Listener Connector SYN No Listener SYN+ACK RST ACK ACKs Data FIN FIN High Port (Usually) Known Port (Usually) High Port (Usually) Known Port (Usually) 3

4 Other Views of TCP Packets In Data In Socket Magic Socket Data Out Packets Out (at link rate) Programmer s View Network View 4

5 Congestion Control TCP controls number of packets in network. Packets are acknowledged, so flow of ACKs. Receiver advertises window to avoid overflow. Congestion window tries to adapt to network. Slow start to roughly find capacity. Congestion avoidance gradually adapts. 5

6 The Congestion Window k'th congestion event w i w i (k+1) w i (k) t a (k) t b (k) t c (k) k'th congestion epoch Time (RTT) Additive increase, multiplicative decrease (AIMD). To fill link need to reach BW τ. Backoff by 1/2, implies buffer is BW τ. Fairness, responsiveness, stability,... 6

7 TCP On Linux Network stack buffers in-flight data. Socket buffer must be BW τ. /proc/net/core/{r,w}mem_max sockbuf sizes. /proc/net/ipv4/tcp_{r,w,}mem min/def/max tcp window. Trade off memory is wired, so valuable. Defaults have recently been increased. 7

8 Research Work Packet loss not caused by congestion. Filling big BW τ product packet at a time. Bad for long-distance high-bandwidth links. Various solutions in pipeline (BIC, Scalable, High-Speed, FAST, H-TCP). Pluggable congestion control in Linux (behind TCP_CONG_ADVANCED). /proc/sys/net/ipv4/tcp_congestion_control Working on other congestion detection techniques. 8

9 Throughput Throughput Flow 1 Throughput Flow Throughput Flow 1 Throughput Flow Throughput (KBytes) Throughput (KBytes) time (sec) time (sec) 9

10 Cwnd 12 Cwnd flow 1 Cwnd flow 2 12 Cwnd flow 1 Cwnd flow Cwnd (packets) 6 Cwnd (packets) time (sec) time (sec) 1

11 Practical Stuff Other issues at play, such as implementation quality. For example ACK processing and queueing problems. Testing is important: land speed records. Project with OSDL to build validation suite. 11

12 Before 25 snd_cwnd snd_nxt throttle 4.5e+9 4e+9 1 slow path 2 3.5e+9 1 Segments (cwnd, ssthresh) e+9 2.5e+9 2e+9 1.5e+9 Bytes (sequence) Distribution e+9 1 5e time (seconds) microsecond 12

13 After 25 2 snd_nxt snd_una snd_cwnd snd_ssthresh qlen throttle 4.5e+9 4e+9 3.5e slow path Segments (cwnd, ssthresh) e+9 2.5e+9 2e+9 1.5e+9 Bytes (sequence) Distribution e+9 1 5e time (seconds) microsecond 13

14 82.11(e) MAC Summary After TX choose rand(, CW 1). Wait until medium idle for DIFS(5µs), While idle count down in slots (2µs). TX when counter gets to, ACK after SIFS (1µs). If ACK then CW = CW min else CW = 2. Ideally produces even distribution of packet TX. In 11e have multiple queues. Each has own CW min, DIFS(aka AIFS) and can have TXOP. 14

15 Testbed setup Multiple STA (Linux) connected to AP (Linux hostap). Hardware model 1 AP Desktop PC 18 STA Soekris 1 STA Desktop PC WLAN NIC Atheros AR5212 External antenna, PCI interface, Madwifi driver with local patches for 11e parameter setting. 15

16 16

17 Validation 5.5 Relative Throughput against TXOP parameter 3 Relative Throughput against AIFS parameter Analytical Model Experimental Results Relative Throughput Relative Throughput TXOP in unit of us AIFS in unit of 2us timeslot Relative Throughput Relative Throughput against CWmin Analytical Model Experimental Results CWmin Shift ( corresponds to the default CWmin 32) Measure relative performance of two saturated flows while varying TXOP, AIFS and CW min. Compare to well-known models. 17

18 Before TCP Uploads to AP for 18 sec 12 6 Uploads and 6 Downloads for 18 sec 1 1 Throughput(kbits/sec) Throughput(kbits/sec) STA id STA id 18

19 After 6 12 TCP Uploads to AP for 18 sec 7 6 Uploads and 6 Downloads 5 6 Throughput(kbits/sec) Throughput(kbits/sec) STA id STA id 19

20 Unprioritised Voice Throughput Delay Unpriotised Station Delay bound for queue stability Throughput (kbps) Average Delay (seconds x 1-6) Unpriotised Station Ideal Throughput 9% Throughput Number of competing stations Number of competing stations 2

21 Measuring Delay Want to measure one-way MAC delay. NTP slow and insufficiently accurate. Simultaneously observable TX better, largish noise..1 station 1 station 2 station 3 station 4.5 Offset (s) Time (s) 21

22 Delay Technique Transmission not complete until MAC ACK. Hardware supports interrupt after ACK. Interface TX Queue 1. Driver notes enqueue time. 2. Hardware contends until ACK received Driver Driver TX Discriptor Queue Hardware 4. Driver notes completion time. 3. Hardware interrupts driver. Packet transmitted ACK received 22

23 Validation Median delay (seconds x 1-6) Long preamble Fit line: 11.12Mbps x us Short preamble Fit line: 11.16Mbps x us Packet size (bytes, excluding headers) 23

24 Fraction of Packets.15.1 Fraction of total packets Delay (seconds x 1-6) Delay (seconds x 1-6) Fraction of total packets.15.1 Fraction of Packets Delay (seconds x 1-6) Delay (seconds x 1-6) 24

25 AIFS Impact Throughput Delay Unpriotised Station Competitors AIFS = 4 Competitors AIFS = 6 Delay bound for queue stability Throughput (kbps) Unpriotised Station 1 Competitors AIFS = 4 Competitors AIFS = 6 Ideal Throughput 9% Throughput Number of competing stations Average Delay (seconds x 1-6) Number of competing stations 25