David Brownell <[EMAIL PROTECTED]> writes:

> Alexander Hoogerhuis wrote:
> > The patch *WITHOUT* debugging enabled fell over real fast. Now *WITH*
> > debugging the skyrocketin load symptom is gone, and it's chugging
> > along nicely at line speed with FTP and also with CPU maxed out with
> > scp about 20% udner line speed (i.e. saturated CPU). SO far I've moved
> > 4Gb of data with no ill effect, which is a new record. I'll let it
> 
> Interesting ... it did have an effect.  Not entirely unexpected,
> though I'm still curious which debug-related delays matter.
> 

I've gone back and played with it, and the sure way to make it go bonk
is to make emacs/gnus look for new mail, at which point I cause
massive I/O to my hard drive (and on a laptop thats sloow).

> 
> > play all night with bouncing files around and see if it keels over at
> > some point. The interrupt rate is roughly 9k/sec for the USB controller.
> 
> That interrupt rate is absurdly high, considering
> 
>   - The controller's irq latency is set at 1 microframe, and
>     there are are only 8000 microframes per second.  Where
>     would the extra IRQs come from?
> 
>   - On 2.5 the scatterlist logic cuts the interrupt rate down
>     to one per i/o scatterlist, instead of one per entry.
>     (The scatterlist below is "big", 127 entries.)
> 
>     At peak rates, EHCI can do about 25 entries per millisecond,
>     many controllers do closer to half that.  Call the "average"
>     sglist about 20 entries (close to what I've noticed), and that
>     suggests a peak IRQ rate closer to 1000/sec.
> 
> Something would seem to be getting measured wrong, or else there's
> something else causing a lot of your load problems.
>

The interrupt rate is higher with debugging enabled, it seems to make
a difference of about 3-4.5k int/sec in difference. I had a few mails
from a guy that ran on RedHat what had three of these drives on his
machine and cooked along nicely, I had a copy of iostat and the likes
from a run of copying against his drives and he came in on the same
numbers as me wihtout debugging.

mvh,
A
-- 
Alexander Hoogerhuis                               | [EMAIL PROTECTED]
CCNP - CCDP - MCNE - CCSE                          | +47 908 21 485
"You have zero privacy anyway. Get over it."  --Scott McNealy


-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to