network performance tests

[Schoenfeld]

Hi,

I'd like to know the network performance of your setups! The tool(s) to use would be NetIO, which is a standard in the networked world. You start the tool on two computers in your home network and have them transfer data packets between each other. The results are printed in a few lines of text like this:

NETIO - Network Throughput Benchmark, Version 1.32
(C) 1997-2012 Kai Uwe Rommel

TCP connection established.
Packet size  1k bytes:  1019.50 KByte/s Tx,  1023.35 KByte/s Rx.
Packet size  2k bytes:  1051.95 KByte/s Tx,  1036.57 KByte/s Rx.
Packet size  4k bytes:  1048.03 KByte/s Tx,  1026.24 KByte/s Rx.
Packet size  8k bytes:  1063.83 KByte/s Tx,  1029.55 KByte/s Rx.
Packet size 16k bytes:  1058.40 KByte/s Tx,  1026.92 KByte/s Rx.
Packet size 32k bytes:  1045.30 KByte/s Tx,  1010.55 KByte/s Rx.
Done.

Please include the type of your system (Amiga, CPU, MHz), the networking card and the stack (plus any tweaks, if any) that you've used. In the case above, it would be:

1. System: A4000/CSPPC 68060RC60
2. card: X-Surf 1, driver version V1.4
3. stack: Genesis (taken from OS3.9) with AmiTCP_PacketFix (Aminet)
4. tweaks: TCP_SendSpace = 16384 and TCP_RecvSpace = 16384


What I am particularly interested in is the performance of PCI-Ethernet solutions such as Mediator, G-Rex and Prometheus, but also PCMCIA ethernet solutions on A1200 and A600.

You can find the Amiga-port of NetIO here:
http://aminet.net/package/comm/tcp/netio132.lha

And other versions for pretty much any OS can be downloaded here: http://www.ars.de/netio

thanks,
Jens

[Jope]

All connections via a twisted pair cable with a switch or two between the machines.

1. System: A4000/CSMK2 68060 50MHz
2. card: Ariadne II, driver version V43.10
3. stack: Roadshow
4. tweaks: Defaults

TCP connection established.
Packet size 1k bytes: 704.64 KByte/s Tx, 850.96 KByte/s Rx.
Packet size 2k bytes: 747.46 KByte/s Tx, 890.07 KByte/s Rx.
Packet size 4k bytes: 782.09 KByte/s Tx, 875.54 KByte/s Rx.
Packet size 8k bytes: 804.11 KByte/s Tx, 882.76 KByte/s Rx.
Packet size 16k bytes: 809.62 KByte/s Tx, 870.85 KByte/s Rx.
Packet size 32k bytes: 815.62 KByte/s Tx, 864.36 KByte/s Rx.

--

1. System: A3000 68030 25MHz 16MB motherboard RAM, 8MB Zorro II RAM
2. card: Ariadne II, driver version V43.10
3. stack: Roadshow
4. tweaks: Defaults

TCP connection established.
Packet size 1k bytes: 324.20 KByte/s Tx, 355.33 KByte/s Rx.
Packet size 2k bytes: 386.09 KByte/s Tx, 400.65 KByte/s Rx.
Packet size 4k bytes: 385.54 KByte/s Tx, 439.30 KByte/s Rx.
Packet size 8k bytes: 392.20 KByte/s Tx, 484.62 KByte/s Rx.
Packet size 16k bytes: 383.19 KByte/s Tx, 521.70 KByte/s Rx.
Packet size 32k bytes: 389.65 KByte/s Tx, 521.83 KByte/s Rx.

--

1. System: A1200/B1260 68060 50MHz
2. card: 3C589D, driver version V1.4
3. stack: Roadshow
4. tweaks: Defaults

TCP connection established.
Packet size 1k bytes: 782.90 KByte/s Tx, 913.34 KByte/s Rx.
Packet size 2k bytes: 800.57 KByte/s Tx, 909.12 KByte/s Rx.
Packet size 4k bytes: 818.94 KByte/s Tx, 926.37 KByte/s Rx.
Packet size 8k bytes: 833.04 KByte/s Tx, 945.73 KByte/s Rx.
Packet size 16k bytes: 763.28 KByte/s Tx, 971.98 KByte/s Rx.
Packet size 32k bytes: 800.81 KByte/s Tx, 935.91 KByte/s Rx.

--

1. System: A1200/GVP1230+ 68030 40MHz
2. card: 3C589D, driver version V1.5
3. stack: Roadshow
4. tweaks: Defaults

TCP connection established.
Packet size 1k bytes: 379.47 KByte/s Tx, 424.97 KByte/s Rx.
Packet size 2k bytes: 449.55 KByte/s Tx, 490.45 KByte/s Rx.
Packet size 4k bytes: 469.15 KByte/s Tx, 543.70 KByte/s Rx.
Packet size 8k bytes: 484.33 KByte/s Tx, 606.08 KByte/s Rx.
Packet size 16k bytes: 474.94 KByte/s Tx, 664.62 KByte/s Rx.
Packet size 32k bytes: 482.95 KByte/s Tx, 652.97 KByte/s Rx.

[Antti]

1. System: A1200/B1230 Mk-IV 68030 50MHz
2. card: D-Link DFE-670TXD, cnet.device V1.9
3. stack: Roadshow
4. tweaks: none

NETIO - Network Throughput Benchmark, Version 1.32
(C) 1997-2012 Kai Uwe Rommel

TCP connection established.
Packet size  1k bytes:  439.10 KByte/s Tx,  556.80 KByte/s Rx.
Packet size  2k bytes:  515.55 KByte/s Tx,  652.44 KByte/s Rx.
Packet size  4k bytes:  548.25 KByte/s Tx,  669.32 KByte/s Rx.
Packet size  8k bytes:  548.86 KByte/s Tx,  702.60 KByte/s Rx.
Packet size 16k bytes:  554.64 KByte/s Tx,  748.92 KByte/s Rx.
Packet size 32k bytes:  549.08 KByte/s Tx,  737.12 KByte/s Rx.
Done.

[matthey]

NETIO - Network Throughput Benchmark, Version 1.32
(C) 1997-2012 Kai Uwe Rommel

TCP connection established.
Packet size  1k bytes:  2033.78 KByte/s Tx,  321.98 KByte/s Rx.
Packet size  2k bytes:  405.09 KByte/s Tx,  574.51 KByte/s Rx.
Packet size  4k bytes:  723.33 KByte/s Tx,  766.37 KByte/s Rx.
Packet size  8k bytes:  1320.06 KByte/s Tx,  876.72 KByte/s Rx.
Packet size 16k bytes:  1618.56 KByte/s Tx,  1356.06 KByte/s Rx.
Packet size 32k bytes:  1644.96 KByte/s Tx,  1663.01 KByte/s Rx.
Done.

Note: Each number can vary several hundred.

1. System: 3000T CSMK3 68060@75MHz 50ns SIMMs
2. card: Mediator with 100MBit/s ethernet (Voodoo 4 mem buffer)
3. stack: Genesis (taken from OS3.9) with AmiTCP_PacketFix (Aminet)
4. tweaks: TCP_SendSpace = 16384 and TCP_RecvSpace = 16384

[thebajaguy]

Late to the party (only 4+yrs) but...<br />
<br />
<br />
NETIO - Network Throughput Benchmark, Version 1.32<br />
(C) 1997-2012 Kai Uwe Rommel<br />
<br />
TCP connection established.<br />
Packet size 1k bytes: 1501.87 KByte/s Tx, 1987.91 KByte/s Rx.<br />
Packet size 2k bytes: 1612.60 KByte/s Tx, 2252.57 KByte/s Rx.<br />
Packet size 4k bytes: 1878.65 KByte/s Tx, 2208.22 KByte/s Rx.<br />
Packet size 8k bytes: 1.95 MByte/s Tx, 2259.73 KByte/s Rx.<br />
Packet size 16k bytes: 2.06 MByte/s Tx, 2260.65 KByte/s Rx.<br />
Packet size 32k bytes: 2084.23 KByte/s Tx, 2272.48 KByte/s Rx.<br />
Done.<br />
<br />
System: A4000T, TekMagic 68060/60Mhz/64M, 16MB Mobo, BigRAM 256MB<br />
Card: X-Surf-100/Z3 100Mbit, using the 68000-coded 1.11 SANA driver (I crashed all the others at some point)<br />
Stack: AmiTCP 3.0Beta <br />
Tweaks: None

Edit: I've since crashed all of the drivers, and as such don't think it's that, but more likely the AmiTCP b3 - I have to put in Pro (v4) soon so will update when I get there.

[Thorham]

1. System: A1200 Blizzard 1230 MK-IV 50mhz
2. Card: PCMCIA 3c589.device compatible
3. Stack: AmiTcp 3.0
4. Tweaks: None that I know of.

NETIO - Network Throughput Benchmark, Version 1.32
(C) 1997-2012 Kai Uwe Rommel

TCP connection established.
Packet size  1k bytes:  427.18 KByte/s Tx,  452.64 KByte/s Rx.
Packet size  2k bytes:  470.77 KByte/s Tx,  547.62 KByte/s Rx.
Packet size  4k bytes:  515.80 KByte/s Tx,  590.68 KByte/s Rx.
Packet size  8k bytes:  537.25 KByte/s Tx,  668.97 KByte/s Rx.
Packet size 16k bytes:  564.18 KByte/s Tx,  681.81 KByte/s Rx.
Packet size 32k bytes:  504.32 KByte/s Tx,  683.07 KByte/s Rx.
Done.

Software can be downloaded from github: https://github.com/kai-uwe-rommel/ne...eases/tag/1.33
Amiga software also on Aminet (includes usage example in the readme): http://aminet.net/package/comm/tcp/netio

[patrik]

Originally Posted by Thorham

4. Tweaks: None that I know of.

Tweaks that usually makes a noticeable difference to network performance is kickstart in fast via blizzard maprom and exec.library library base in fast. Especially the latter does not happen automatically as the Blizzard does not add its memory via normal autoconfig.

[Thorham]

No one interested in adding some benchmark data?

Originally Posted by patrik

Tweaks that usually makes a noticeable difference to network performance is kickstart in fast via blizzard maprom and exec.library library base in fast

Very interesting. Perhaps I"ll check it out.

[patrik]

1. System: A1200 Blizzard 1260 50mhz (MapROM enabled), Kickstart 3.2.1, Workbench 3.2.1
2. Card: PCMCIA Netgear FA411 100Mbit using cnet16.device
3. Stack: Roadshow 1.14
4. Tweaks: No tweaks so exec.library library base, system stack and interrupt vectors are in slow chip

7.Ram Disk:> residentspeed 
Type     Name                    Version Size    Loc  Speed   Speed %
Region   interrupt vector table            1.00k Chip   3.68M    10.8
Region   system stack                      6.00k Chip   3.65M    10.5
LibBase  expansion.library        47.3       554 Chip   3.68M    10.3
Resident 1234-BootPrefs0           0.0        42 Chip   3.86M    14.2
LibBase  exec.library             47.8     1.50k Chip   3.68M    10.6
7.Ram Disk:> netio -t 192.168.1.1 
 
NETIO - Network Throughput Benchmark, Version 1.32
(C) 1997-2012 Kai Uwe Rommel

TCP connection established.
Packet size  1k bytes:  413.16 KByte/s Tx,  609.01 KByte/s Rx.
Packet size  2k bytes:  461.64 KByte/s Tx,  722.58 KByte/s Rx.
Packet size  4k bytes:  596.24 KByte/s Tx,  737.80 KByte/s Rx.
Packet size  8k bytes:  655.07 KByte/s Tx,  923.77 KByte/s Rx.
Packet size 16k bytes:  0.67 MByte/s Tx,  934.49 KByte/s Rx.
Packet size 32k bytes:  0.69 MByte/s Tx,  939.10 KByte/s Rx.
Done.

4. Tweaks: MuMove4k + MuFastZero ON FASTEXEC FASTSSP to move exec.library library base, system stack and interrupt vectors to fast. (FastExec is an alternative to the MuTools to achieve this and on kick 3.1 BlizKick and its LocalFast module is the best)

7.Ram Disk:> residentspeed 
Type     Name                    Version Size    Loc  Speed   Speed %
Resident 1234-BootPrefs0           0.0        42 Chip   3.38M    12.5
Resident « MuMove4K »             40.23    1.67k Chip   3.65M    10.6
7.Ram Disk:> netio -t 192.168.1.1 
 
NETIO - Network Throughput Benchmark, Version 1.32
(C) 1997-2012 Kai Uwe Rommel

TCP connection established.
Packet size  1k bytes:  642.28 KByte/s Tx,  864.23 KByte/s Rx.
Packet size  2k bytes:  712.20 KByte/s Tx,  998.14 KByte/s Rx.
Packet size  4k bytes:  857.09 KByte/s Tx,  1010.39 KByte/s Rx.
Packet size  8k bytes:  918.05 KByte/s Tx,  1184.72 KByte/s Rx.
Packet size 16k bytes:  0.94 MByte/s Tx,  1221.02 KByte/s Rx.
Packet size 32k bytes:  0.96 MByte/s Tx,  1228.95 KByte/s Rx.
Done.

[Thorham]

Wow, that's quite a difference, didn't expect it to be that much

[jkdsteve]

I assume the fastest transfer rates completely max out the CPU? Impacting system usability

[patrik]

The CPU is the bottleneck yes, but many of its cycles are also spent in unproductive ways on waitstates to access the network card.

Anyway, the system is generally perfectly usable as both Roadshow and the app transferring data (ttcp in this case) runs at default priority of 0. If another task needs cpu, it will get its fair share.

[Olaf Barthel]

Originally Posted by patrik

The CPU is the bottleneck yes, but many of its cycles are also spent in unproductive ways on waitstates to access the network card.

Anyway, the system is generally perfectly usable as both Roadshow and the app transferring data (ttcp in this case) runs at default priority of 0. If another task needs cpu, it will get its fair share.

Roadshow will let the caller keep its Task/Process priority, so you could tune a program's priority to get a larger or smaller share of the CPU. You can also change the controller Process' priority with the RoadshowControl command as needed.

AmiTCP does this differently. The controller Process raises its priority to 5 when AmiTCP starts, if I remember correctly, and each caller's priority will be changed to the same priority which the controller Process uses. This may imply a Task rescheduling.