Re: NetBSD hardclock overhead growth

[John Klos <john%ziaspace.com@localhost>]

> > With the binary you sent, and 100.000 iterations:
> > Microseconds for one run through Dhrystone:  103.6
> > Dhrystones per Second:                      9652.5
>
>  Ok, this is very strange.  I dug out one of my VLC systems and running the 
> 5.1 pkgsrc version of dhrystone gives me:
> Microseconds for one run through Dhrystone:   89.9
> Dhrystones per Second:                      11123.5
>
> 8.8u 0.1s 0:10.89 82.4% 0+0k 0+0io 0pf+0w
> # sysctl hw.model
> hw.model = VAXstation 4000 VLC
> # uname -v
> NetBSD 7.99.19 (GENERIC) #0: Sun Aug 23 14:45:02 MDT 2015 
> mhitch@nuc.hitch:/home/mhitch/NetBSD-current/OBJ/vax/home/mhitch/NetBSD-current/src/sys/arch/vax/compile/GENERIC
>
>  I don't know why John gets such poor results.
>
>  I wonder if the ROM version would make any difference, if there is some 
> difference in how the system is initialized before booting.
>
>  This is from the VLC I ran the above on:
>
> KA48-A V1.2-32B-V4.0
> 08-00-2B-23-02-4B
> 24MB

I have the 1.3 ROM. There are two obvious possibilities:

The 1.2 ROM initializes the cache tags properly. This doesn't seem likely 
because the notes which talked about the upgrade from 1.2 to 1.3 didn't 
mention anything about the cache.

Or... my cache has a defective page (or two, or three) whereas yours does 
not, so my machine's cache doesn't get enabled because the code is meant 
for the KA660 and doesn't take in to account bad pages.

I'll have to do some tests to see if I can get my cache enabled. Yay!

John