Sidebar
Table of Contents
Connectivity tests via Singapore
Default Oz-Europe route is via Singapore and 4×155 Mbps links (aggregated). This has the lowest RTT which is what most users want.
Testing Methology
iperf in TCP and UDP mode. In TCP mode trying various different window sizes. Most tests were short, 30-120 sec
Hosts
| Host | Location |
|---|---|
| epvsi1-ext | Sydney |
| pkvsi1-ext | Parkes |
| cavsi1-ext | Narrabri |
| mpvsi1-ext | Mopra |
| Huygens | Netherlands |
UDP Tests
Sydney → Netherlands (9/1/2007)
| Nominal Rate | Achieved Average | Packet Loss |
|---|---|---|
| 10 Mbps | 9.9 Mbps | 1[=%=] |
| 50 Mbps | 50 Mbps | 0[=%=] |
| 100 Mbps | 100 Mbps | 0.02[=%=] |
| 120 Mbps | 120 Mbps | 0.25[=%=] |
| 130 Mbps | 126 Mbps | 3.5[=%=] |
| 200 Mbps | 136 Mbps | 33[=%=] |
Sydney → Netherlands (10/1/2007)
| Nominal Rate | Achieved Average | Packet Loss |
|---|---|---|
| 100 Mbps | 100 Mbps | 0.03[=%=] |
| 120 Mbps | 118 Mbps | 1.4[=%=] |
Observatories → Netherlands (10/1/2007)
| Site | Nominal Rate | Achieved Average | Packet Loss |
|---|---|---|---|
| Parkes | 100 Mbps | 100 Mbps | 0.000[=%=] |
| Parkes | 120 Mbps | 118 Mbps | 1.6[=%=] |
| Narrabri | 100 Mbps | 100 Mbps | 0.37[=%=] |
| Narrabri | 120 Mbps | 117 Mbps | 2.5[=%=] |
| Mopra | 100 Mbps | 100 Mbps | 0.39[=%=] |
| Mopra | 120 Mbps | 116 Mbps | 3.5[=%=] |
Simultaneous Tests (Multiple UDP streams from observatories, 11/1/2007)
Running three iperf sessions simultaneously (120sec each)
| Sites | Nominal Rate | Achieved Average | Packet Loss |
|---|---|---|---|
| Parkes+Mopra | 100 Mbps | 100/100 Mbps | 0.17/0.17[=%=] |
| Parkes+Narrabri | 100 Mbps | 100/100 Mbps | 0.3/0.3[=%=] |
| Sydney,Sydney,Sydney | 100 Mbps | 49/48/46 Mbps | 51/51/54[=%=] |
| Mopra+Narrabri+Parkes | 100 Mbps | 47/47/47 Mbps | 53/53/53[=%=] |
15/1/2007
It turns out links are shared on a hashing algorithm based on IP address and port numbers. This is deterministic but unlucky choice of port number may mean two streams are forced onto the same 155 Mbps link even though this leads to asymmetric use.
Testing with careful choice of port number…
| Sites | Port | Nominal Rate | Achieved Average | Packet Loss |
|---|---|---|---|---|
| ATCA+Mopra | 52102/52100 | 100 Mbps | 100/100 Mbps | 0.39/0.39[=%=] |
TCP tests
These tests with iperf with various window sizes. Given the relatively slow ramp up time the average and ~95 percentile values are given (eyeballed).
Round trip time is 311 msec.
60 second tests
| Sites | TCP window | Average | Peak |
|---|---|---|---|
| Sydney | 2 MB | 14 Mbps | 18 Mbps |
| Sydney | 1 MB | 16 Mbps | 20 Mbps |
| Sydney | 800 kB | 12 Mbps | 14 Mbps |
| JIVE | 4 MB | 23 Mbps | 24 Mbps |
| JIVE | 5 MB | 40 Mbps | 43 Mbps |
| JIVE | 2 MB | 43 Mbps | 48 Mbps |
| JIVE | 1 MB | 19 Mbps | 20 Mbps |
| JIVE | 3 MB | 56 Mbps | 65 Mbps |
| Sydney | 2 MB | 29 Mbps | 36 Mbps |
| Sydney | 3 MB | 44 Mbps | 60 Mbps |
