[LinuxPPS] More Results.

[clemens at dwf.com]

Over the last 2=3 days I have run linuxPPS bis and ter and ter+IRQ
with ntp 4.2.4p4 and 4.2.5p113 in various combinations.

First, everything seems to work fine.

Second, at least from my diagnostics, there is not as much of an improvement
in adding the IRQ patch as I had expected.

There are two issues.
The ONCORE driver prints the times of the 'timehacks' (there has got to be a
better term) to the clockstats file.  I plot difference between the timehack 
and
the second, vs time.

What one sees is a 'fuzzy' wavy line.
Its not clear what causes the low frequency (wavy) part of the line, but the
fuzzy part is clearly just the difference in time it takes the kernel to answer
the interrupts.

This FUZZ does seem to be SLIGHTLY narrower with the IRQ change
than without it, but I had expected a major change and this is perhaps 25%.

The thing that IS DIFFERENT is the outriders.
In addition to the fuzz and the wavey parts of the line, there are individual
points that lie 10 to 50us above the curve.
These are clearly delays in getting the timehack when some other interrupt
was in process.

There are perhaps an ORDER OF MAGNITUDE less of these large
outriders when using the IRQ patch.

This is obviously nice to see, tho I expect that ntp would toss these outriders
in its normal 'cleanup' of the timehacks before using them.

---

So thats where I stand.
The IRQ patch does reduce the fuzz, but not as much as I had expected.
And it does reduce the number of outriders by an order of magnitude.

I can put out an email with a couple of plots if anyone is interested,
but the Postscript plots I generate tend to be large.

-- 
                                        Reg.Clemens
                                        reg at dwf.com