Thankyou for all your replies. I am combining this into one big re-reply.
What's the impedance of your multimeter?
It is an UT61E, it did not list frequency measurement but norminal impedance for AC is 10MOhm. The frequency measurement is an "ALT Key" triggered function. It was in parallel with the USB-Scope. The scope impedance is 1MOhm norminal. So 10M//1M, overall load impedance would be 900K-ish.
When the scope ran alone, I noted no error in an overnight run (4 to 5 million 56 byte reads) under the same condition. Perhaps when it got pull further down, it was just enough. When it is done with this round of torture test, I will put an 800K resistor as load soldered on (no dangling wire as long antenna) and see how that goes.
I think that, since you have an oscilloscope, the place to start your learning exercise is to use it to look at what happens to the signal when you connect your DMM to it.
Already did, nothing exceeding the normal scope noise. The scope is the Hantek USB 6022BE - a rather noise one to begin with.
Did you put pull-up resistors on the i2c bus ? How 'hard' are they ? They may be too soft... Put 1k
A pair of 3K pull up for SDA/SCL.
Just a few thoughts...
1) You will need error handling in the software anyway.
2) You don't know if the error was because of the DMM connected, a lost neutrino, a barking dog or whatever.
3) How often are you going to read? One error in 1 million when you read every second is 11.5 days.
...
(1)
I plan to add software error handling in the codes.
But I feel if I just "cheated" by ignored the error. When doing it merely for fun, I don't want to feel I didn't do a complete job. Thus I want to know from more experience hands if this kind of error rate is an expected thing or not. (I didn't find much real-world data on I2C error rate research including reading a PDF from Phillips, the inventor of I2C.)
(2) Dog (next door) did bark... Would I2C error be good justification to shoot the dog? A side thought - the darn "test-bed" is resting 6 inches away from an Laserjet 5P and some printing did occur. I know from prior work it affected my ATMEGA doing ADC volt-reading.
(3) I have no plan on using the clock in any demanding role if any at all. That is not the point tho. As a learning exercise, I just want to see if my work is passing the smell test as reasonable work.
It's not really necessary to care about effects of test equipment on the circuit you're designing. There's a reason why high input impedances on voltage-measuring devices and low ones on current measuring are preferred.
I kind of thought that. Particularly when I have no plans on using the square wave nor would I want to continue to verify what frequency it is at. But
I was trying to get a measure of the "quality of my work". So I was thinking along the line - if I had buy one that was professionally designed and built, would it be as "error prone". I think 3.3million read and just one error, it is "ok to use" for me. I like to think of it as good work, but came to realize I really don't know how to judge.
But my goal is a learning exercise. If it could be better, I want to give it a good try. What do you think? Is that kind of error something to be expected? -or- if I should do a better job?
...
Thanks, all! From your comments, I get the sense that it is below par. I think I will go back to tinkering with it some more to find a way of allowing it to hook up to the scope and my DMM and see if I can get it to complete at least 10million read-56bytes test without error. (That would be a full day's worth of read/compare.) See if I get there and then see where I go from there.
I think I need to modify the "test" by first "quantify the noise" I inject. "A dangling DMM probe wire" really is uncontrolled. I need more thought there. Good time to think a bit more since my soldering iron died right after I completed that PCB. I ordered a temperature controlled one but until the replacement iron arrives, there is not much else I can do but to think about it...
Thanks again! I welcome more thought and advice.
Looking forward to testing myself to see if I can beat that problem (or see how far I can decrease the error rate...)