CCID-3 over CSLIP
This document describes using CCID-3 over Compressed Serial Line IP (CSLIP, RFC 1144).1. Preliminaries
The serial ports of two computers are connected via a Null-Modem cable. Useful background information can be found in the Serial-HOWTO.The kernel needs to have both SLIP and CSLIP enabled (under Network device support).
2. Setting up the IP-over-serial-line link
The
program slattach is used
to create an
interface on each host's
serial port:slattach -vd -s 115200 -p cslip /dev/ttyS0
This creates an interface, sl0, which needs to be
configured. First it is enabled via
ip link set dev sl0 up
Then a point-to-point
link is set up as
follows:
Testing with ping verifies that this connection works. The interface has a small MTU (296) and queue length (10):
| On the first computer: | ip addr add 10.0.0.1/24 peer 10.0.0.2 dev sl0 |
| On the second computer: | ip addr add 10.0.0.2/24 peer 10.0.0.1 dev sl0 |
Testing with ping verifies that this connection works. The interface has a small MTU (296) and queue length (10):
% ip -s link show sl0
6: sl0: <POINTOPOINT,MULTICAST,NOARP,UP,10000> mtu 296 qdisc pfifo_fast qlen 10
link/cslip
RX: bytes packets errors dropped overrun mcast
47416 893 0 0 0 0
TX: bytes packets errors dropped carrier collsns
639188 2338 0 0 0 1
3. Speed considerations
Before doing any DCCP tests, it is useful to estimate the maximum link speed. The maximum line speed depends on the type of serial port. To find out the maximum line speed of the serial port(s), try setserial -a -g /dev/ttyS* for standard-PC serial ports and /dev/ttyUSB* for USB serial converters. For the standard 16550 UART in PCs this is 115200 bps, USB serial port converters can do higher speeds.Bytes sent out via the serial port are framed using a start-bit and a stop-bit, so that when parity bits are not used the maximum link speed is 80% of the line speed. The following table shows measured link speeds when no parity is used.
| Line
Speed in kbps |
2.400 |
9.600 |
19.200 |
38.400 |
57.600 |
115.200 |
|---|---|---|---|---|---|---|
| Link
Speed in kbps |
1.6 |
6.4 |
12.8 |
25.5 |
38.3 |
82.0 |
| Ratio | 66.7% |
66.7% |
66.7% | 66.4% | 66.5% | 71.2% |
| RTT in msec | 734 |
185 |
94 |
48 |
33 |
17.7 |
These speeds were determined using iperf with TCP cubic over a 60-second test run each. RTT values are averages of 10 ping measurements each.
The ratio Link-speed/Line-speed shows that additional overhead reduced the maximum link speed available to TCP from 80% down to approximately 66.7% (2/3). This overhead includes the framing used by the SLIP protocol (RFC 1055) plus other overhead, minus the reductions afforded by compressing the TCP/IP headers using CSLIP (RFC 1144).
The following table shows the impact of performing header compression, and compares against the 6-bit mode of SLIP where only a subset of the ASCII protocol ("printable SLIP") is used.
| Line
Speed in kbps |
2.400 |
9.600 |
19.200 |
38.400 |
57.600 |
115.200 |
|---|---|---|---|---|---|---|
| CSLIP
/ kbps |
1.6 |
6.4 |
12.8 |
25.5 |
38.3 |
82.0 |
| SLIP
/ kbps |
1.6 |
6.3 |
12.6 |
25.2 |
37.8 |
75.6 |
| SLIP6
/ kbps |
1.2 |
4.7 |
9.5 |
18.9 |
28.4 |
56.7 |
For the selected speed ranges, the gains of using CSLIP are not very high. Since CSLIP compresses TCP headers and not DCCP headers, the benefit of using CSLIP over using SLIP is even less for DCCP.
Together with start/stop bits, the throughput utilisation of SLIP6 can not be more than 6/10 = 60%. In the table, the actual value was close to 49% in each case.
4. DCCP test runs with iperf
For the DCCP tests, CSLIP was used with a line speed of 115.2 kbps. The following commands were used for the test run, taking into account the smaller MTU (and hence MPS).| On the first computer: | iperf
-sdb -l 236 |
| On the second computer: | iperf
-dc 10.0.01 -l 236 -t 10 |
The receiver
reported a throughput of
73.1 Kbps over 10 seconds, which agrees with the asymptotic value of
the plot below.
Due to the small queue length, 30
packets were dropped at the
sender.
The following diagram shows the allowed
sending rate computed for the above-given Loss Rate p.
The link RTT measured with ping was 17 msec on average.
The CCID-3 RTT below shows a tendency to "climb".
To better follow this trend, the diagram
below shows the evolution of the RTT over a longer period.
