Serial/Ethernet Converter

Ask the poeple who sell it.

A former employer of mine has a whole line of products that do what you want. They devices range from models with 1 serial port to models with 32 serials ports. From the host end, the ports can be accessed via TCP sockets or you can install a device driver for Linux or Windows that make the ports accessible as "normal" serial ports (/dev/ttyXX under Linux or COMXX under Windows).

Check out the Lantronix Xport

I have recently worked with B&B's ESO901 serial server. It has also the virtual serial port (Windows) but it seems they have latency problems (like 250ms on the serial loopback), so I had to write the code which does access the serial port using sockets. B&B's technical support on this issue was close to none, but I found out they licensed the product from Passport Networks

may be u might get interested ?

This message was sent using the comp.arch.embedded web interface o

We have used the moxa ne-4110s successfully

Ross

That seems to be a pretty common problem. When I worked at Comtrol, we put a fair bit of effort in to keeping latency to a minimum, and our products were an order of magnitude faster the some of the competition's devices. Properly configured, our remote serial ports could have lower latency than a native

Here's a web page with a cute flash drag race animation showing the latency testing results:

There's a table of results on the page, and a link to download a whitepaper explaining the testing methodology and the benchmark program.

Excuse my ignorance, but how can you achieve "better" latency than "native" serial port ? On local serial port the latency is as low as latency of driver stack (if any), unlike remote serial port will have latency of: driver stack + network transport + serial server processing.

The data may be innacurate, especialy 5.35ms for the local serial port. I achieved oscilloscope measurable ~0.2ms response time on Win2k system on one project with ConnectTech BlueHeat and disabled FIFO.

Maybe the 5ms on local port was measured with the card not transmitting immediately if FIFO is not full.

On 2005-08-18, rziak wrote:\

There's latency (mostly receive) in the UART -- especially if the FIFO is enabled. For some cases the latency in the UART exceeds the network latency. You can minimize the latency in the UART by disabling the FIFO, but then you've got an awful lot of interrupts to service.

Disabling the FIFO is the key. IIRC, our tests were all done with the FIFO enabled because that's how our customers run their native ports -- I presume because they've had problems running native ports in one interrupt per byte mode.

The latency on the local port is mostly in the receive FIFO. If the recieve data stops, there's a timer that has to timeout before a receive interrupt is generated.