Re: static binary timings (Was: Some more patches for GCC on NetBSD/VAX coming soon...)

[David Brownlee <abs%netbsd.org@localhost>]

On 30 March 2016 at 12:24, Johnny Billquist <bqt%update.uu.se@localhost> wrote:
> On 2016-03-30 13:15, Anders Magnusson wrote:
>>
>> Den 2016-03-30 kl. 10:09, skrev David Brownlee:
>>>
>>> One thing you may find is that shared libraries have a significant
>>> performance penalty over static linking on VAX
>>
>> It would be interesting to do some real timing measurements here, since
>> the vax assembler itself is virtually PIC and shouldn't give any
>> performance
>> penalty.
>
>
> Someone did a year or two ago, and showed that the penalty was small. I'll
> try to locate the post later today.

I did a test on an emulated VAX, then found and issue and retested
with a follow up email, but oddly the follow up does not appear to be
in the list archives...

Included below:


---------- Forwarded message ----------
From: David Brownlee <abs%netbsd.org@localhost>
Date: 19 May 2014 at 00:31
Subject: NetBSD/vax simh startup times (pic,static,nopam,makemandb!)
To: port-vax List <port-vax%netbsd.org@localhost>


It turns out I borked my previous static/dynamic timing tests (Bad
David, Bad, Bad!)

So, have run some more tests in NetBSD/vax simh, tested more, and
definitely some interesting results.

Once again everything should be taken with a strong pinch of salt as
this on an emulated VAX on a much faster box.

Three builds
- Default, ra90 image
- MKPAM=no, ra90 image
- NOPIC=yes, ra92 image (would not fit on ra90)

Two actions timed)
- time from final number on kernel boot line to login prompt (+/-1 sec
variation)
- time from RETURN on login to # prompt

Results
- Default: 2:45 14
- NOPAM: 2:44 10
- NOPIC: 2:32 21

The NOPIC number in particular was a little odd - faster to boot but
slower to login?

It just so happened that earlier today someone on port-mac68k@ had
noticed that makemandb was running for a long time after booting on
their machine, making it close to unusable. A quick check, and yes
'makemandb -Q' was running. makemandb updates the manpage index,
stat()ing every man page on the system. Nothing to worry about on a
fast box, but somewhat unreasonable to run on every boot on a slow on.

A quick 'echo makemandb=NO >> /etc/rc.conf' and re-run the tests

New Results
- Default: 2:02 6
- NOPAM: 2:02 4
- NOPIC: 1:36 3

Around 1 second of the login time is processing terminal
initialisation string - setting terminal to 'dumb' dropped the NOPAM
number to 3 seconds.

Conclusions (albeit with simh timings)
- having makemandb running makes timings (and the machine) pretty much useless
- NOPAM reduced login time by 33%
- NOPIC reduced login time by 50%
- NOPIC reduced boot time by 25%, even allowing for fixed times such
as kernel boot

I'm going to take the makemandb issue off into another thread -
leaving the PAM/PIC timings here for comment :)

Does anyone have a genuine VAX to hand they could see if makemandb is
running after boot, and if so to try 'echo makemandb=NO >>
/etc/rc.conf' (& reboot)?

Emulator notes
- simh 3.9.0 on a Thinkpad T420s
- machdep.est.frequency.target=800 (its slowest speed) to maximize differences

Build information
- NetBSD/vax build from current 6.99.41
- All builds MKX11=no
- Each emulated system booted once, default console terminal set to
xterm, then shutdown
- simh config & build images are uploading to
http://sync.absd.org/vax/timing-images/