Client
|
Server
|
Network
|
Throughput (KB/s)
|
Frame rate (fps)
|
Application Latency (ms)
|
Transmission Time (s)
|
Completed Frames
|
Partial Frames
|
Partial Frame %
|
Ping
|
|---|
3GP4-MdkLnx | Loki | 1000 Mb/s Smart Switch | 6450 +/- 675 | 8.7 +/- 0.9 | 1322 +/- 666 | 0.117 +/- 0.014 | 1 +/- 0 | n.a | n.a. | 0.62 +/- 0.88 |
3GP4-WinXPSP2 | Loki | 1000 Mb/s Smart Switch | 6665 +/- 386 | 9.0 +/- 0.5 | 1294 +/- 357 | 0.112 +/- 0.007 | 1 +/- 0 | n.a. | n.a. | 0.0 |
PBG4 | Loki | 1000 Mb/s Smart Switch | 4060 +/- 757 | 5.47 +/- 1.02 | 55.0 +/- 24.2 | 0.563 +/- 0.067 | 1.8 +/- 0.48 | 2.2 +/- 1.8 | 37.2 +/- 7.7 | 0.62 +/- 0.48 |
PBG4 | Loki | 100 Mb/s Drexel LAN | 8530 +/- 131 | 11.5 +/- 1.75 | 49.7 +/- 5.77 | 0.424 +/- 0.070 | 3 +/- 1 | 4.7 +/- 1.5 | 42% +/- 9.8 | 0.77 +/- 0.55 |
PBG4 | Loki | 1.5 Mb/s DSL-East coast | 141 +/- 2 | 0.20 +/- 0.0 | 103 +/- 32.3 | 6.17 +/- 0.100 | 1 +/- 0 | 8.0 +/- 4.2 | 2.37 +/- 0.92 | 27 +/- 15 |
PBG4 | NrtrnBIRN | 1.5 Mb/s DSL-East coast | 104 +/- 17 | 0.15 +/- 0.02 | 238 +/- 122 | 8.12 +/- 1.36 | 1 +/- 0 | 5.8 +/- 1.8 | 1.89 +/- 0.55 | 163 +/- 123 |
PBG4 | NrtrnBIRN | 100 Mb/s Drexel LAN | 752 +/- 116 | 1.0 +/- 0.16 | 104 +/- 12.4 | 1.22 +/- 0.195 | 1 +/- 0 | 2.3 +/- 0.58 | 8.9% +/- 0.44 | 74 +/- 3.0 |
- 20 identical knife slice requests were sent in rapid succession (inter-request interval < 0.1 ms). Requested cross-sections consisted of 576 × 320 (184320 pixels). Were all 20 returned in their entirety to the client, they would consist of ~740 separate 249 pixel rasters. Those rasters returned by the server were not rendered in order to control for client display performance. Metrics for both throughput and latency are the mean of 5 independent tests run ~10 seconds apart. As indicated, not all requested images were returned in full due to the efficiency logic in the server and client. The metrics, based on those pixel rasters returned to the client, are: throughput(total byes/total time excluding application latency), frame rate(throughput scaled to test image size), application latency(time in ms from first request to first raster returned), transmission time(time in seconds from first to last raster returned for all 20 requests), completed frames(number of complete image raster sets returned), partial frames(number of incomplete image raster sets returned), partial frame %(for incomplete images the percent rasters returned), ping(round-trip latency to server). For all of the tests where client and server were on separate machines, one of two NtAS servers were used (Loki at Drexel U., Philadelphia, PA, USA or NrtrnBIRN at UCSD, San Diego, CA, USA), each of which had the following configuration: dual 2.3 GHz G5 PPC Macintosh XServe/2.0 GB PC3200 ECC SDRAM (SBus 667 MHz, Cache Bus 1.15 GHz) running Mac OS X 10.3.9. The following clients configurations were tested: PBG4 (Macintosh PowerBook 1.5 GHz G4 PPC/1.5 GB PC2700 DDR SDRAM running Mac OS X 10.4.8); 3GP4-MdkLnx (3.0 GHz Pentium 4/2 GB DDR RAM running Mandrake-Linux v10.1); 3GP4-WinXPSP2 (3.0 GHz Pentium 4/1 GB DDR SDRAM running Windows XP SP2 w/2GB VM). All clients machines used the JVM v1.5.0 to run NetOStat – for Linux and Windows XP as distributed by Sun Microsystems; for Mac OS X as distributed by Apple Computer. All tests were performed at 3 AM EST, when both the Internet and LAN activity is at a minimum and is the most stable.
