I have some older measurements here: https://www.brutman.com/mTCP/mTCP_Performance.html

I'll refresh them sometime before the summer is over, but I don't expect the numbers to change too much. I'm also interesting in getting reports from other people with different hardware.

The benchmark measures pure TCP/IP performance. All of the standard checksumming and retries are done, but there are no buffer copies - once the TCP packet has been received and processed the data is thrown away. This gives you an idea of the limits of your machine. These benchmarks are not difficult to run, but you do have to have access to Linux or something similar on the other side to be able to run Netcat.

mbbrutman wrote on Yesterday, 20:05:

https://www.brutman.com/mTCP/mTCP_Performance.html

I'm trying the procedure described there, but:

1$ ncat --version
2Ncat: Version 7.92 ( https://nmap.org/ncat )
3
4$ time nc -N 192.168.1.102 2000 < bogus.bin
5nc: invalid option -- 'N'
6Ncat: Try `--help' or man(1) ncat for more information, usage options and help. QUITTING.

What's the equivalent of "-N" in current ncat?

Edit:
This doesn't work, either:

1$ time nc -d 192.168.1.102 2000 > bogus.bin
2Ncat: Invalid -d delay "192.168.1.102" (must be greater than 0 and less than 2147483s). QUITTING.

Grzyb wrote on Yesterday, 19:14:

Does the 3c5X9CFG utility work on 8088 ?
If it does, then show your card's config!

With pleasure:

I've got new BETTER results, new infos coming up in a few minutes.

I suspect that you have duplex mismatch...

The card is set to Full Duplex, but doesn't support NWAY, therefore it can't negotiate it with the switch.
Have you manually set the port on the switch to FD?
If not, then it works in Half Duplex.

FD vs. HD mismatch can reduce performance a lot.
I don't know how much it matters with an 8088, but it definitely doesn't help.

mbbrutman wrote on Yesterday, 20:05:

I have some older measurements here: https://www.brutman.com/mTCP/mTCP_Performance.html

I'll refresh them sometime before the summer is over, but I don't expect the numbers to change too much. I'm also interesting in getting reports from other people with different hardware.

The benchmark measures pure TCP/IP performance. All of the standard checksumming and retries are done, but there are no buffer copies - once the TCP packet has been received and processed the data is thrown away. This gives you an idea of the limits of your machine. These benchmarks are not difficult to run, but you do have to have access to Linux or something similar on the other side to be able to run Netcat.

Thank you for these great programs, the least I can do is support with data. 😀

Previous run with FTP; the main PC was running over WiFi. Now both PCs are connected with LAN.

New results below.

I was lazy to do anything linux now, so I used windows.
Installed netcat from: https://nmap.org/download.html
CMD used:
Send: echo %TIME% && ncat 192.168.1.73 2000 > bogus.bin && echo %TIME%
The time shown is odd (same time), command terminated when 4MB sent.

Receive: echo %TIME% && ncat -w 60 192.168.1.73 2000 < bogus.bin && echo %TIME%
The time shown is odd (same time), command did not terminate so was running forever, added 60s stop command.
PS. Maybe I was just impatient... the result seems to be finished after the 4MB transfer is complete: within 60seconds.

Send 1:53,394
Send 2:53,399
Send 3:53,394
Bytes sent: 4194304
Average send speed: 76.71 KB/s

Receive 1: 52,290
Receive 2:53,300
Receive 3:52,294
Bytes received: 4194304
Average receive speed: 77.83 KB/s

Images

Send:

Receive:

Grzyb wrote on Yesterday, 21:51:

Have you manually set the port on the switch to FD?

Did nothing on the router/switch.
Funnily the IBM XT is connected to a 4 gigabit port FTTH (ZTE ZXHN F668) router. 😀
(main PC connected to the same router)

There is not much I can do on the router side.
I choose this LAN card for the XT because my previous WD8003W (8bit ISA) LAN card could not negotiate with this router.
More details on this in the previous posts.

JAKra85 wrote on Yesterday, 22:02:

There is not much I can do on the router side.

In this case you probably should disable Full Duplex on the card.

Grzyb wrote on Yesterday, 22:31:

JAKra85 wrote on Yesterday, 22:02:

There is not much I can do on the router side.

In this case you probably should disable Full Duplex on the card.

If I do that it will fail to connect.
This router can't do Half-Duplex. 😁
The reason I switched from Half-Duplex WD8003W to Full-Duplex 3C509B.

I will try it though tomorrow, just to be 100% sure.

However, I think these speeds are quite acceptable:
Send speed: 76.71 KB/s
Receive speed: 77.83 KB/s

Grzyb wrote on Yesterday, 22:31:

JAKra85 wrote on Yesterday, 22:02:

There is not much I can do on the router side.

In this case you probably should disable Full Duplex on the card.

I must correct myself.:
It works in Half-Duplex mode too.
I do not know why the WD8003W (8bit ISA) LAN was not working, not important anymore.
(missing auto-negotiation feature maybe?)

New windows command, with fixed timings:
Send: cmd /v:on /c "echo %TIME% && ncat 192.168.1.73 2000 > bogus.bin && echo !TIME!"
Receive: cmd /v:on /c "echo %TIME% && ncat 192.168.1.73 2000 < bogus.bin && echo !TIME!"

New measurement below.
IBM XT (AMD 8088 CPU@4.77MHz), running IBM DOS v3.3, using 4MB file.

Half-Duplex:
Send: 75.17 KB/s
Receive: 76.82 KB/s

Full-Duplex:
Send: 76.70 KB/s
Receive: 76.75 KB/s

8mb/sec is pretty close to the maximum that you can squeeze out of an ISA bus, IIRC. (The theoretical max bandwidth of a 16bit ISA bus is 8.3mb/sec. Realworld will have other stuff crossing the bus, and will also observe waitstates, or CPU cycle misses.)

JAKra85 wrote on Today, 08:05:

I do not know why the WD8003W (8bit ISA) LAN was not working, not important anymore.
(missing auto-negotiation feature maybe?)

From what I know, Nway Auto-Negotiation was introduced with 100 Mbps.
I'm not aware of any 10 Mbps card with Nway support.

Anyway, it doesn't seem important with such a slow CPU.
Overall, your results look perfectly normal.

DaveDDS wrote on Yesterday, 16:09:

FWIW, if you want to test ethernet with the absolute minimum of local processing overhead.... Try transferring a file with my DD […]
Show full quote

JAKra85 wrote on Yesterday, 13:21:

After a bit of research I think my 44KB/s with 3C509B (16bit in 8bit mode) is on the low side. ...

FWIW, if you want to test ethernet with the absolute minimum of local processing overhead....
Try transferring a file with my DDLINK - DDLINK uses a very low level protocol in "raw" packets.
I designed it to be very small to include with ImageDisk on a bootable floppy, hence it doesn't do IP(FTP), or SMP etc.
- I do have a tiny/simple stand-alone TFTP as well

DDLINK will run on an 8088 - I don't have one now with networking, but I use it to move files to/from an ancient 8088 "Poqet PC" over serial frequently.

I just looked, and there is a 3C509 packet driver in the collection I publish in various bootable disks.

Transferring a file, disk speed is probably going to affect you more than Ethernet speed - There are RAMdrives that work on an 8088, but they have to fit with DOS in 640k so they are small (but can be big enough for some simple testing - fortunately DDLINK.COM is only about 17k)

Thank you, the goal has been achieved.
For "daily" tasks I always tend to use what is simplest and the one I am already familiar with.
Currently copying files to and from the XT, on the XT I just simply run "ftpsrv" command and that's it.
On the Win11 I start WinSCP, connect and start copying.