GR8FTP bug report thread | MSX Resource Center (Pagina 9/10)

Van Manuel

Ascended (19463)

19-08-2018, 17:24

Hmm, I'm surprised about the low speed of the DenYoNet... Aren't there known specs about its speed?

Van Louthrax

Prophet (2465)

19-08-2018, 17:31

Have not found any specs yet (and am suprised too !).

BTW, I'd be interested if somebody has a PDF or scanned version of the DenYoNet manual.

Van fr3nd

Expert (112)

19-08-2018, 17:51

I've tested it with a Tecnobytes Ethernet card (Obsonet2 compatible) on a Turbo-R and it seems to work properly!

Van Louthrax

Prophet (2465)

19-08-2018, 17:58

Ah, nice news! That's where some software standardization comes handy

I'm curious about the download rate of your card, if you can (even roughly) measure it?

Van fr3nd

Expert (112)

19-08-2018, 18:14

I just downloaded a 128K file in 28seconds approximately...

Van Grauw

Ascended (10768)

19-08-2018, 22:15

Bit disappointed with UNAPI transfer speed… Of course it’s an abstraction layer, but since they are block memory operations I wouldn’t expect it to slow things down so much…

Van Louthrax

Prophet (2465)

19-08-2018, 22:44

Grauw wrote:

Bit disappointed with UNAPI transfer speed… Of course it’s an abstraction layer, but since they are block memory operations I wouldn’t expect it to slow things down so much…

For the GR8Net, my guess is that the current UNAPI implementation is just not optimized yet, there's probably lots of room for improvement.

For the DenYoNet, not sure that the slowness is related to the UNAPI layer: on turboR, there's almost no gain in using the DenYoNet/UNAPI in R800 mode compared to Z80, so that looks more hardware-related...

Van Louthrax

Prophet (2465)

19-08-2018, 22:48

fr3nd wrote:

I just downloaded a 128K file in 28seconds approximately...

So only 4.5KB per second (2x slower than the DenYoNet)? Ah well, it's still a bit faster than a tape

Van Eugeny_Brychkov

Paragon (1232)

11-10-2018, 15:59

Louthrax wrote:

For the GR8Net, my guess is that the current UNAPI implementation is just not optimized yet, there's probably lots of room for improvement.

API calls are always slower than direct access to the hardware. One of the reasons is that UNAPI runs in CPU bank 2 (8000-BFFF), thus loads data to temp location first, then switches it out to previous slot location in bank 2, and moves data to required location. No doubt slot switching and two data moves instead one are not a quick tasks for every transaction of max 2 KB. As routine is universal firmware call, it can not be fast by design...
I got to know about these performance issues surfing the forum, do not recall anyone informed me about them.
I do not know how application works, can not advise. If you prefer let's have private chat so that we may get an idea how to increase the performance.

Van edoz

Prophet (2482)

11-10-2018, 19:51

Maybe it is related to the w5100. Seems it is not the fastest chip. But I'm not a hardware specialist. What I notice is that the denyonet is sometimes not reliable as well when downloading bigger files. Sometimes it just waits doing nothing and then suddenly it continues.. I heard from Yobi that this this could be a hardware issue as well. But still not sure. At least the buffering part of the w5100 could sometime have some strange behaviours as well.I did a lot of tests in the past when I did test work for symbos, and it always came out that the gr8net is very stable compare to the denoyonet. Don't get me wrong. Both are superior hardware devices for me and I still glad they exists. Next to that, lot of network protocols have a lot of info inside which need to be processed by the z80. On the other hand, if I look to the gr8net cloud solution im very surprised by the speed.