Bug in 1.94 with 38k
#11
I started writing some functional tests, and ran into some odd behavior that doesn't seem to be related to the original problem. It seems that if you have bi-directional transfers going on (on the internet side), that can lead to some very long delays, and also loss of connection and hang (requiring power cycle).

The test goes something like this...

1. Connect to CS and enter P2P mode. In this state, every character sent will be echoed back directly, a sort of loop-back state.
2. Send the string "test,test,test,test"
3. Capture all the characters coming back, recording a timestamp (jiffies) when pulling out of the rx buffer.

Here's what happens... Occasionally, the whole string comes back instantly. But more often, there is a 3 to 5 second delay before the first byte comes back. After the 5th byte, there is another delay of 3 to 4 seconds, after which the remainder of the string comes back. If this is repeated enough times (usually 5 to 10 times), then the WiModem stops responding, and eventually goes from "connected" to "Ready". But when this happens, it will not respond to commands anymore. Whether I use the 2400 baud driver or the 38k driver, I just get back occasional garbage characters and I have to power cycle to regain control.

Shorter strings come in with no delay, and if I switch back and forth between Rx and Tx after every few bytes, it seems to work fine. It's just when I send more than about 15 bytes without asserting RTS and reading any incoming data.
Reply
#12
This occurs at 2400 baud as well?
Reply
#13
Only tested with 38k so far. I'll test 2400 baud next to see what happens there.
Reply
#14
Yes, problem also occurs at 2400 baud. Was kinda surprised. Main difference I'm seeing with 2400 baud is that the first character sent is getting dropped. Not sure if that's a problem with my test, but there are very similar delays. Sometimes the first character doesn't get dropped. Sometimes the delays aren't there... sometimes only one delay rather than 2. The first character was corrupted once, rather than missing.

I'll try to get the test programs polished up a bit tomorrow and send them to you.
Reply
#15
Ok, thanks!
Reply




Users browsing this thread: 3 Guest(s)