Data preparation¶
IO format¶
All algorithm input and output are defined in csv format.
Input:¶
Following are the stream set and network description files as algorithm input.
Stream set format:
stream |
src |
dst |
size |
period |
deadline |
jitter |
|---|---|---|---|---|---|---|
0 |
0 |
[7,8,9] |
50 |
100000 |
44600 |
0 |
stream: Unique ID for each flow
src: Talker that the end-system where flow starts at.
dst: Listener that the the end-system where flow ends at, formatted as list for multicast
size: Packet size of each flow in Bytes.
period: Flow periods in Nanoseconds.
deadline: Relative flow deadline requirement in Nanoseconds.
jitter: Maximum end-to-end delay variance requirement in Nanoseconds.
Network format:
link |
q_num |
rate |
t_proc |
t_prop |
|---|---|---|---|---|
(0, 1) |
8 |
1 |
1000 |
0 |
link: Directional link connects two devices.
q_num: Number of available queues for time-triggered (critical) traffic.
rate: Bandwidth of link in bit / nanosecond, e.g., 1 = 1 Gbps, 0.1 = 100 Mbps, 0.01 = 10 Mbps.
t_proc: Processing time including switching fabric and ingress processing.
t_prop: Propagation delay on wire after transmission.
Nodes:
Each device is a node that has directional links to other devices. End stations (talkers/listeners) are nodes that have two directional links connecting it to a bridge. Each stream should begin and end at an end station node. Switches (or bridges) are nodes that connect to two or more other nodes. In the stream file, the source nodes and destination nodes should not be switches.
Output¶
Following are the output files (gcl, offset, route, queuing assignment) from the algorithm, which can be fed into the TSN simulator or testbed. The implementation can be found in tsnkit.core._config.py
GCL:
link |
queue |
start |
end |
cycle |
|---|---|---|---|---|
(0, 1) |
0 |
1000 |
5000 |
12000000 |
queue: Indicator implies which queue is open between start and end time.
start: Relative time when queue opens in hyper period.
end: Relative time when queue opens in hyper period.
cycle: Cycle time of GCL.
Offset:
stream |
frame |
offset |
|---|---|---|
0 |
0 |
1000 |
stream: Unique ID for each stream.
frame: The index of corresponding flow instance
offset: The traffic dispatching time on end-system for corresponding flow instance
Route:
stream |
link |
|---|---|
0 |
(8,1) |
0 |
(1,2) |
0 |
(2,3) |
link: Directional link connects two devices.
Queueing assignment:
stream |
frame |
link |
queue |
|---|---|---|---|
0 |
0 |
(8,1) |
2 |
stream: Unique ID for each flow.
frame: The index of corresponding flow instance.
link: Directional link connects two devices.
queue: The egress queue for corresponding flow instance on corresponding link.
Data generator¶
The data generator is a python script that generates random stream set and network description files. The script is located at data/generator.py. The script takes the following arguments:
–num_ins: Number of problem instances to generate, default is 1.
–num_stream: Number of flows in each problem instance, default is 8.
–num_sw: Number of network bridge in each problem instance, default is 8.
–period: Period pattern of stream set.
–size: Size pattern of stream set.
–deadline: Deadline pattern of stream set.
–topo: Topology pattern of network.
–output: Output directory for generated files.
To generate a single dataset:
python3.10 -m tsnkit.data.generator --num_ins 1 --num_stream 8 --num_sw 8 --period 1 --size 1 --deadline 1 --topo 1
To generate multiple datasets:
## Generate 6 datasets with combinations of [4,8,18] flows, [1,2] payload, and repeat 2 times. (Total 6*2=12)
python3.10 -m tsnkit.data.generator --num_ins 2 --num_stream 4,8,18 --num_sw 8 --period 1 --size 1,2 --deadline 1 --topo 1
## Output:
## Generating dataset - ins0: 100%|███████████████████████████████████| 6/6 [00:00<00:00, 360.29it/s]
## Generating dataset - ins1: 100%|███████████████████████████████████| 6/6 [00:00<00:00, 405.10it/s]
In the output file dataset_logs.csv:
stream,size,period,deadline,topo,num_stream,num_sw
1,1,1,1,1,4,8
2,2,1,1,1,4,8
3,1,1,1,1,8,8
4,2,1,1,1,8,8
5,1,1,1,1,18,8
6,2,1,1,1,18,8
7,1,1,1,1,4,8
8,2,1,1,1,4,8
9,1,1,1,1,8,8
10,2,1,1,1,8,8
11,1,1,1,1,18,8
12,2,1,1,1,18,8
The specific size/deadline/period/topology patterns can be found in src/tsnkit/data/dataset_spec.py or in our paper.