1. Benchmarking SpaceWire Networks
International SpaceWire Conference 2007
17-19 September 2007
Benchmarking SpaceWire Networks
Asaf Baron, Isask'har Walter, Ran Ginosar, Isaac Keslassy
EE Department, Technion, Haifa, Israel
Ofer Lapid
Israel Ministry of Defense
General Notes:
Use the word Space Craft and not Satellite
Do not use shortcuts (NCI, PLP) when speak

2. The problem Multiple proposals for SpaceWire
Vendors design new SpaceWire parts
BUT …
How can we evaluate proposals?
How can we compare parts from different vendors?

3. Benchmark

4. Benchmark purposes Evaluate protocol proposals
Compare network configurations
Cost
Performance
Reliability
Compare components
Formalize requirements
Improve the standard

5. Benchmark Contents End units Traffic specifications
Redundancy
Traffic specifications
Unit X sends to unit Y
Quality of Service Requirements
Bandwidth, real time constraints, deadlines, Priorities, etc.
Benchmark does NOT include implementation details
No topology, no specific components

6. End units (examples) # Active Units # Units Unit
Data Handling System (DHS) 2 1
Reconfiguration Unit (RU)
Payload Sensor (Camera) 3 1
Downlink 2
Solar Cells Control 2 1
In this slide we should mention the word Redundancy
Star Tracker 3

7. Traffic matrix
Most important…

8. Traffic matrix (examples)
High bandwidth
From To
Type BW
B/sec
Latency
Recommended
Priority
Sensor
Storage
Payload
100M L
Low bandwidth
DHS
Sensor
Control
1000
Low H

9. Using the Benchmark

10. Evaluating SpaceWire features
Standard: Packet-level priority (PLP)
Non-Standard: N-Char interleaving (NCI)

11. N-Char Interleaving Not in SpaceWire standard
But, Very useful in Wormhole networks
Requires multiple VCs
VC code added to each N-Char
Ideally, SPW would allow longer flits
Specify that we do a short break with the example to explain about NCI.
Ideally, SPW would allow longer flits when NCI is used since otherwise the overhead bits per N-Char might be too big

12. No-NCI Example 46 44 40 48 42 52 58 56 54 38 36 24 22 26 28 34 32 30 60 50 10 8 6 4 62 14 20 18 16 2 12 68 66 64 70 72 78 76 74 80
100 Mb/s
Delay
Red: 26
Yellow: 48
Green: 69
400 Mb/s
Router #1
Router #2
Fast link is under (not fully) utilized
400 Mb/s

13. NCI Example (4 VC)
55.2
52.8 48
57.6
50.4
62.4
69.6
67.2
64.8
45.6
43.2
28.8
26.4
31.2
33.6
40.8
38.4 36 72 60 12
9.6
7.2
4.8
74.4
16.8 24
21.6
19.2
2.4
14.4 84
81.6
79.2 96
86.4
93.6
91.2
88.8
76.8
Delay
100 Mb/s
Red (26): 31
Yellow (48): 38
Green (69): 45
400 Mb/s
Router #1
Router #2
Change same as in No-NCI
400 Mb/s

14. Network topology downlink router sensor storage
DHS
storage
router
sensor
Only active units shown and simulated
Redundant units, routers, links hidden and ignored

15. Simulator display Simulated using OPNET
A commercial simulator for networks

16. Simulations: Low priority payload traffic
1. SIMPLE NET
4. NCI+PLP
2. ADD PLP
5. NCI+PLP, more VCs
3. ADD NCI
First explain what we see.
When I talk about the x-axis say “traffic load”.
Ignore the reduced ETE delay when we are using priority.
When we used NCI only we got that the network can transmit 3 times more PLD traffic than what it could transmit when NCI was not used

17. Simulations Results – Other Priorities
High Priority
Medium Priority
NCI with PLP is beneficial
Low ETE delay of med+high priority packets
High ETE delay of low priority packets
Low ETE delay of med+high priority packets in regardless of the load

18. Summary A new SpaceWire benchmark (open source)
A new simulator for SpaceWire
An example of using the benchmark
Priorities help
Non-standard NCI helps even more