perf

bandwidth / throughput

• bandwidth / data rate: maximum rate we can send per unit time
  • most commonly measuring the speed of a link
• 1 gigabit/second = transmit 1 bit / nanosecond

• throughput: acheived rate per unit time

  • often lower than total bandwidth because of losses
  • (we’ll give several examples throughout the semester)

latency (1)

• latency: time for message: SOURCE \(\rightarrow\) DEST
• example: 1000 bit message from S to D:

• one bit sent each 1/50M second = 0.02 \(\mu\) s
• 1000 bits take \(0.02 \times 1000 = 20\;\mu\) s to sent
  • ‘‘transmission delay’’
• • \(2.2\) microseconds for bit to go down cable (\(2.3\times10^8\) m/s)
  • ‘‘propogation delay’’
• total latency of about 22.2 \(\mu\) s

latency (1, ex)

• exercise: latency for 20000 bit message from S to D
  • assume speed of signal through fiber of \(2.0\times10^8\) m/s

latency (2)

• example: 1000 bit packet from S to D

• assume when message is received:

  • 5 other 1000-bit packets in queue; no extra bits between packets
  • no other switch processing time

• S to switch, switch to D: 22.2 \(\mu\) s (transmit+propogate delay)

• within switch: wait \(20\times5=100\;mu\) s for 5 other packets (\(20\mu s\) = 1 packet transmit delay)

  • ‘‘queueing delay’’

• total latency: \(22.2 + 100 + 22.2 = 144.4\) microseconds

jitter

• variation in latency

  • most commonly from changing queuing delays

Figure 20 from Section 1.5 of Computer Networks: A Systems Approach (6th ed) (Peterson and Davie)

measuring round-trip time (1a)

charles@reisst14$ ping 1.1.1.1
PING 1.1.1.1 (1.1.1.1) 56(84) bytes of data.
64 bytes from 1.1.1.1: icmp_seq=1 ttl=52 time=13.8 ms
64 bytes from 1.1.1.1: icmp_seq=2 ttl=52 time=15.0 ms
64 bytes from 1.1.1.1: icmp_seq=3 ttl=52 time=12.5 ms
64 bytes from 1.1.1.1: icmp_seq=4 ttl=52 time=12.3 ms
64 bytes from 1.1.1.1: icmp_seq=5 ttl=52 time=13.5 ms
64 bytes from 1.1.1.1: icmp_seq=6 ttl=52 time=12.5 ms
64 bytes from 1.1.1.1: icmp_seq=7 ttl=52 time=13.3 ms
64 bytes from 1.1.1.1: icmp_seq=8 ttl=52 time=13.2 ms
64 bytes from 1.1.1.1: icmp_seq=9 ttl=52 time=13.3 ms
64 bytes from 1.1.1.1: icmp_seq=10 ttl=52 time=14.1 ms
^C
--- 1.1.1.1 ping statistics ---
10 packets transmitted, 10 received, 0% packet loss, time 9014ms
rtt min/avg/max/mdev = 12.273/13.343/15.024/0.786 ms

non-ICMP pings (1)

HPING www (enp0s31f6 128.143.67.8): NO FLAGS are set, 40 headers + 0 data bytes                          
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=0 win=0 rtt=3.5 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=1 win=0 rtt=3.2 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=2 win=0 rtt=7.1 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=3 win=0 rtt=6.8 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=4 win=0 rtt=6.5 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=5 win=0 rtt=6.2 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=6 win=0 rtt=5.8 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=7 win=0 rtt=5.4 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=8 win=0 rtt=5.0 ms                            
len=46 ip=128.143.67.8 ttl=63 DF id=0 sport=0 flags=RA seq=9 win=0 rtt=4.7 ms                            
^C                                                                                                       
--- www hping statistic ---                                                                              
10 packets transmitted, 10 packets received, 0% packet loss                                              
round-trip min/avg/max = 3.2/5.4/7.1 ms

measuring throughput?

$ scp test.dat portal.cs.virginia.edu:test.dat
test.dat                         100%   32MB  23.0MB/s   00:01    
$ scp portal.cs.virginia.edu:test.dat .
test.dat                         100%   32MB  28.2MB/s   00:01   

• (but might be measuring disk speed instead)

• also more specialized tools like iperf
  • require program to run on both ends

measuring throughput

$ iperf -s
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size:  128 KByte (default)
------------------------------------------------------------
[  1] local 128.143.71.87 port 5001 connected with 128.143.71.27 port 54760
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-10.0147 sec  1.09 GBytes   934 Mbits/sec

$ iperf -c kytos02 | tee iperf.out
------------------------------------------------------------
Client connecting to kytos02, TCP port 5001
TCP window size: 85.0 KByte (default)
------------------------------------------------------------
[  1] local 128.143.71.27 port 54760 connected with 128.143.71.87 port 5001
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-10.0256 sec  1.09 GBytes   933 Mbits/sec

measuring transmission delay?

$ ping -s 1000 www -i 0.05 -c 1000 -q
PING www.cs.virginia.edu (128.143.67.8) 1400(1428) bytes of data.
--- www.cs.virginia.edu ping statistics ---
1000 packets transmitted, 1000 received, 0% packet loss, time 50638ms
rtt min/avg/max/mdev = 0.319/0.461/1.222/0.039 ms
$ ping -s 16 www -i 0.05 -c 1000 -q
PING www.cs.virginia.edu (128.143.67.8) 16(44) bytes of data.
--- www.cs.virginia.edu ping statistics ---
1000 packets transmitted, 1000 received, 0% packet loss, time 50995ms
rtt min/avg/max/mdev = 0.156/0.345/1.539/0.068 ms

• approx. \(0.461 - 0.345 = 0.116\) ms delay for \(1400-16\) extra bytes

  • with two links in each direction = approx \(\frac{0.116}{4}=0.029\) ms/link
  • \(\frac{1400-16 \text{byte}}{0.029 \text{ms}} \approx 50\) Mbit/sec (does not match Gigabit ethernet)
  • probably other processing time besides sending on links, though