demik
Well-known member
Hello everyone,
One item on my todo list was to benchmark Apple NuBus A/ROSE Ethernet card. A/ROSE is a small Operating System that that runs on some NuBus cards. A/ROSE NuBus cards have their on 68k CPU (usually a 68000 @ 10 MHz) that should offload the main CPU on a Mac II.
People reported hat the A/ROSE card was superior in performance and that the computer felt snappier with this card in it. But how much is it ?
Each runs consisted of uploading the folder to a server and downloading it again.
For each run, I rebooted the Client computer and did the test multiples times, and averaged the results.
All tests were done using stock 7.5.5 drivers and AppleShare Clients, using the AppleTalk Protocol (no TCP/IP used)
The Apple Ethernet card was compared to a more classic Asanté Ethernet card
Both client and servers were directly connected to an entreprise grade network switch with ports speed set to auto
- The server was connected with a 100 Mbps full duplex link
- The client was connected will a 10 Mbps half duplex link
Macintosh IIcx
CPU: 68030 @ 16 MHz
RAM: 8 MB
Virtual Memory: on
HDD: 36GB SCA 15k RPM SCSI Drive
OS: 7.5.5
CPU: G3 @ 500 MHz
RAM: 640 MB
HDD: 36 GB UW 10k RPM SCSI Drive (on Adaptec 2940UW)
OS: Debian 8 + Custom 4.14.222 kernel + Netatalk 2.5.5 + AppleTalk stack
NIC: 3Com Corporation 3c905B "Cyclone"
Apple Ethernet NB Card
Part number: 820-0417-C
ROM version: 341-0246
Build date: 1991
Card 2
Asanté Technologies MC3NB
Part number: MC3NB(ST-NIC-V) Rev B1
ROM version: 000094-04E0D8 024-09-A
Build date: 1991
Note: ROM has been dumped here
OT: Mac OS is tuning with the OpenTransport network stack
Classic: Mac OS is running with the Classic network stack
The A/ROSE Ethernet card is slightly slower than the Asanté one on average.
What is more interesting though, is that OpenTransport doesn't play well on CPU constrained Macs. Both cards are noticeably faster without OpenTransport.
The A/ROSE card is also slower in 24 bits mode
While the A/ROSE card is slower here, during the benchmark I indeed noticed that the screen redrawing was faster during Network Access. Maybe the result are different in others Macs, or Network Servers were the background CPU usage is higher. What's possible is that the CPU usage for the A/ROSE card is lower but the overall latency might be higher (your network data has to go through two CPUs), although I don't know how to check this. It may be interesting to check this in another Mac just for science
Final conclusion
If you want more performance, the network card doesn't matter, the network stack does.
One item on my todo list was to benchmark Apple NuBus A/ROSE Ethernet card. A/ROSE is a small Operating System that that runs on some NuBus cards. A/ROSE NuBus cards have their on 68k CPU (usually a 68000 @ 10 MHz) that should offload the main CPU on a Mac II.
People reported hat the A/ROSE card was superior in performance and that the computer felt snappier with this card in it. But how much is it ?
Benchmark
The benchmark files set consists of 37 random files (Application, SimpleText and AppleGuide files, etc…) for a total size of about 3 MB.Each runs consisted of uploading the folder to a server and downloading it again.
For each run, I rebooted the Client computer and did the test multiples times, and averaged the results.
All tests were done using stock 7.5.5 drivers and AppleShare Clients, using the AppleTalk Protocol (no TCP/IP used)
The Apple Ethernet card was compared to a more classic Asanté Ethernet card
Both client and servers were directly connected to an entreprise grade network switch with ports speed set to auto
- The server was connected with a 100 Mbps full duplex link
- The client was connected will a 10 Mbps half duplex link
Client Configuration
For this benchmark, Mac OS 7.5.3 was reinstalled from scratch on the "client" computer, with updates up to 7.5.5.Macintosh IIcx
CPU: 68030 @ 16 MHz
RAM: 8 MB
Virtual Memory: on
HDD: 36GB SCA 15k RPM SCSI Drive
OS: 7.5.5
Server Configuration
Beige G3 µATX buildCPU: G3 @ 500 MHz
RAM: 640 MB
HDD: 36 GB UW 10k RPM SCSI Drive (on Adaptec 2940UW)
OS: Debian 8 + Custom 4.14.222 kernel + Netatalk 2.5.5 + AppleTalk stack
NIC: 3Com Corporation 3c905B "Cyclone"
Ethernet Cards
Card 1Apple Ethernet NB Card
Part number: 820-0417-C
ROM version: 341-0246
Build date: 1991
Card 2
Asanté Technologies MC3NB
Part number: MC3NB(ST-NIC-V) Rev B1
ROM version: 000094-04E0D8 024-09-A
Build date: 1991
Note: ROM has been dumped here
Table of results
24b / 32b: Mac OS is running in 24bit or 32bit modeOT: Mac OS is tuning with the OpenTransport network stack
Classic: Mac OS is running with the Classic network stack
Test / Benchmark time | Apple A/ROSE | Asanté MC3NB |
24b OT Upload | 60.2 | 55.3 |
24b OT Download | 45.8 | 39.0 |
32b OT Upload | 55.1 | 55.1 |
32b OT Download | 37.1 | 38.8 |
24b Classic Upload | 38.0 | 28.6 |
24b Classic Download | 27.0 | 23.3 |
32b Classic Upload | 32.8 | 28.7 |
32b Classic Download | 23.4 | 23.5 |
Graphs
Times to copy in seconds, lower is betterConclusion
Well hmmm... That's some interesting results...The A/ROSE Ethernet card is slightly slower than the Asanté one on average.
What is more interesting though, is that OpenTransport doesn't play well on CPU constrained Macs. Both cards are noticeably faster without OpenTransport.
The A/ROSE card is also slower in 24 bits mode
While the A/ROSE card is slower here, during the benchmark I indeed noticed that the screen redrawing was faster during Network Access. Maybe the result are different in others Macs, or Network Servers were the background CPU usage is higher. What's possible is that the CPU usage for the A/ROSE card is lower but the overall latency might be higher (your network data has to go through two CPUs), although I don't know how to check this. It may be interesting to check this in another Mac just for science
Final conclusion
If you want more performance, the network card doesn't matter, the network stack does.