Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.misc > #9228 > unrolled thread

Are we just running in place?

Back to article view | Back to comp.misc

Contents

  Are we just running in place? RS Wood <rsw@therandymon.com> - 2015-10-30 15:53 +0300
    Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-10-30 13:43 +0000
      Re: Are we just running in place? Dan Espen <despen@verizon.net> - 2015-10-30 10:28 -0400
        Re: Are we just running in place? Stan Barr <plan.b@bluesomatic.org> - 2015-10-30 15:34 +0000
          Re: Are we just running in place?     wje@acm.org (Bill Evans) - 2015-10-30 09:05 -0700
            Re: Are we just running in place? Dan Espen <despen@verizon.net> - 2015-10-30 13:25 -0400
              Re: Are we just running in place? Bob Eager <news0005@eager.cx> - 2015-10-30 17:28 +0000
              Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-10-30 18:32 +0000
                Re: Are we just running in place? Jorgen Grahn <grahn+nntp@snipabacken.se> - 2015-11-01 20:01 +0000
                  Re: Are we just running in place? Paul Sture <nospam@sture.ch> - 2015-11-04 10:11 +0100
                    Re: Are we just running in place? scott@alfter.diespammersdie.us (Scott Alfter) - 2015-11-04 22:29 +0000
                      Re: Are we just running in place? Larry Sheldon <lfsheldon@gmail.com> - 2015-11-04 16:44 -0600
                      Re: Are we just running in place? Dan Espen <despen@verizon.net> - 2015-11-04 23:04 -0500
                        Re: Are we just running in place? Larry Sheldon <lfsheldon@gmail.com> - 2015-11-04 22:34 -0600
                        Re: Are we just running in place? RS Wood <rsw@therandymon.com> - 2015-11-05 10:30 +0300
                        Re: Are we just running in place? Stephen Chadfield <stephen@chadfield.com> - 2015-11-05 13:39 +0000
            Re: Are we just running in place? Bob Eager <news0005@eager.cx> - 2015-10-30 17:30 +0000
              Re: Are we just running in place? Stan Barr <plan.b@bluesomatic.org> - 2015-10-31 08:05 +0000
          Re: Are we just running in place? Paul Sture <nospam@sture.ch> - 2015-11-04 10:01 +0100
        Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-10-30 17:34 +0000
          Re: Are we just running in place? Dan Espen <despen@verizon.net> - 2015-10-30 13:49 -0400
            Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-10-30 18:23 +0000
              Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-10-30 18:59 +0000
              Re: Are we just running in place? Mike Spencer <mds@bogus.nodomain.nowhere> - 2015-11-01 01:33 -0300
              Re: Are we just running in place? Paul Sture <nospam@sture.ch> - 2015-11-04 10:19 +0100
                Re: Are we just running in place? Bob Eager <news0005@eager.cx> - 2015-11-04 21:21 +0000
          Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-10-30 18:33 +0000
            Re: Are we just running in place? Uncle Steve <stevet810@gmail.com> - 2015-11-01 12:59 -0500
              Re: Are we just running in place? scott@alfter.diespammersdie.us (Scott Alfter) - 2015-11-03 17:29 +0000
                Re: Are we just running in place? fmassei@gmail.com - 2015-11-03 10:14 -0800
          Re: Are we just running in place? Rich <rich@example.invalid> - 2015-10-30 18:39 +0000
            Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-10-30 19:22 +0000
              Re: Are we just running in place? "Dirk T. Verbeek" <dverbeek@xs4all.nl> - 2015-10-30 20:27 +0100
                Re: Are we just running in place? bde@besplex.bde.org (Bruce Evans) - 2015-10-30 20:08 +0000
                  Re: Are we just running in place? Richard Kettlewell <rjk@greenend.org.uk> - 2015-10-30 21:21 +0000
                    Re: Are we just running in place? bde@besplex.bde.org (Bruce Evans) - 2015-10-31 04:15 +0000
                  Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-11-02 14:19 +0000
                    Re: Are we just running in place? bde@besplex.bde.org (Bruce Evans) - 2015-11-02 16:15 +0000
                      Re: Are we just running in place? Dan Espen <despen@verizon.net> - 2015-11-02 12:29 -0500
                      Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-11-02 19:13 +0000
                        Re: Are we just running in place? bde@besplex.bde.org (Bruce Evans) - 2015-11-02 21:25 +0000
                          Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-11-03 14:03 +0000
              Re: Are we just running in place? Rich <rich@example.invalid> - 2015-10-30 21:11 +0000
          Re: Are we just running in place? Mike Spencer <mds@bogus.nodomain.nowhere> - 2015-11-01 01:23 -0300
        Re: Are we just running in place? RS Wood <rsw@therandymon.com> - 2015-10-30 20:05 +0000
      Re: Are we just running in place? Johnny B Good <johnny-b-good@invalid.ntlworld.com> - 2015-10-30 19:53 +0000
        Re: Are we just running in place? Uncle Steve <stevet810@gmail.com> - 2015-11-01 12:09 -0500
      Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-11-01 09:42 +0000
        Re: Are we just running in place? jmfbahciv <See.above@aol.com> - 2015-11-01 13:33 +0000
          Re: Are we just running in place? "Rod Speed" <rod.speed.aaa@gmail.com> - 2015-11-02 12:45 +1100
            Re: Are we just running in place? Morten Reistad <first@last.name.invalid> - 2015-11-02 11:06 +0100
      Re: Are we just running in place? Waldek Hebisch <hebisch@math.uni.wroc.pl> - 2015-11-03 22:06 +0000
        Re: Are we just running in place? Morten Reistad <first@last.name.invalid> - 2015-11-03 23:43 +0100
          Re: Are we just running in place? jmfbahciv <See.above@aol.com> - 2015-11-04 13:52 +0000
        Re: Are we just running in place? BartC <bc@freeuk.com> - 2015-11-03 23:46 +0000
          Re: Are we just running in place? Waldek Hebisch <hebisch@math.uni.wroc.pl> - 2015-11-04 04:35 +0000
            Re: Are we just running in place? Morten Reistad <first@last.name.invalid> - 2015-11-04 10:24 +0100
              Re: Are we just running in place? Ahem A Rivet's Shot <steveo@eircom.net> - 2015-11-04 09:55 +0000
                Re: Are we just running in place? Morten Reistad <first@last.name.invalid> - 2015-11-04 12:42 +0100
            Re: Are we just running in place? BartC <bc@freeuk.com> - 2015-11-04 14:57 +0000
              Re: Are we just running in place? Dan Espen <despen@verizon.net> - 2015-11-04 11:29 -0500
              Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-11-04 16:43 +0000
              Re: Are we just running in place? Morten Reistad <first@last.name.invalid> - 2015-11-04 17:55 +0100
                Re: Are we just running in place? BartC <bc@freeuk.com> - 2015-11-04 18:04 +0000
                  Re: Are we just running in place? scott@slp53.sl.home (Scott Lurndal) - 2015-11-04 18:48 +0000
                    Re: Are we just running in place? "Osmium" <r124c4u102@comcast.net> - 2015-11-04 13:05 -0600
                    Re: Are we just running in place? BartC <bc@freeuk.com> - 2015-11-04 19:36 +0000
                  Re: Are we just running in place? Morten Reistad <first@last.name.invalid> - 2015-11-04 21:15 +0100
                    Re: Are we just running in place? BartC <bc@freeuk.com> - 2015-11-04 21:35 +0000
              Re: Are we just running in place? Kees Nuyt <k.nuyt@nospam.demon.nl> - 2015-11-04 21:07 +0100
          Re: Are we just running in place? Richard Kettlewell <rjk@greenend.org.uk> - 2015-11-04 10:36 +0000
        Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-11-03 23:52 +0000
          Re: Are we just running in place? Andrew Swallow <am.swallow@btinternet.com> - 2015-11-04 02:56 +0000
            Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-11-04 11:05 +0000
          Re: Are we just running in place? Waldek Hebisch <hebisch@math.uni.wroc.pl> - 2015-11-04 03:12 +0000
            Re: Are we just running in place? Morten Reistad <first@last.name.invalid> - 2015-11-04 10:02 +0100
              Re: Are we just running in place? Andreas Eder <a_eder_muc@web.de> - 2015-11-08 11:48 +0100
          Re: Are we just running in place? Paul Sture <nospam@sture.ch> - 2015-11-04 10:36 +0100
          Re: Are we just running in place? "John Jackson" <jj@nospam.com> - 2015-11-05 09:01 +1100
            Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-11-05 00:09 +0000
        Re: Are we just running in place? Peter Flass <peter_flass@yahoo.com> - 2015-11-04 11:03 -0500
          Re: Are we just running in place? Bob Eager <news0005@eager.cx> - 2015-11-04 21:25 +0000
          Re: Are we just running in place? Stan Barr <plan.b@bluesomatic.org> - 2015-11-05 08:01 +0000
    Re: Are we just running in place? "78lp" <78lp@nospam.com> - 2015-10-31 15:09 +1100
    Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-10-31 11:11 +0000
      Re: Are we just running in place? RS Wood <rsw@therandymon.com> - 2015-10-31 14:57 +0300
      Re: Are we just running in place? Dan Espen <despen@verizon.net> - 2015-10-31 10:19 -0400
        Re: Are we just running in place? "gareth" <no.spam@thank.you.invalid> - 2015-10-31 16:28 +0000
      Re: Are we just running in place?     wje@acm.org (Bill Evans) - 2015-10-31 04:14 -0700

Page 3 of 5 — ← Prev page 1 2 [3] 4 5  Next page →

#9340

In article <rBOZx.13293$9Q.2428@fx01.iad>,
Scott Lurndal <slp53@pacbell.net> wrote:
> bde@besplex.bde.org (Bruce Evans) writes:
> >In article <giKZx.6499$CP7.3911@fx11.iad>,
> >Scott Lurndal <slp53@pacbell.net> wrote:
> >> bde@besplex.bde.org (Bruce Evans) writes:
> >> >In article <5633c494$0$23758$e4fe514c@news.xs4all.nl>,
> >> >Dirk T. Verbeek <dverbeek@xs4all.nl> wrote:
> >> >> Op 30-10-15 om 20:22 schreef gareth:
> >> >> > "Rich" <rich@example.invalid> wrote in message
> >> >> > news:n10dgn$uk2$2@dont-email.me...
> >> >> >> The fvwm2 executable is 862,808 bytes
> >> >> >
> >> >> > Why?
> >> >> >
> >> >> Because it is from another era?
> >> >
> >> >If it were from another era, then it wouldn't be so bloated:
> >> >
> >> >-r-xr-xr-x  1 root  wheel  124616 Jan  4  2005 fvwm95*
> >> >-r-xr-sr-x  1 root  kmem   403716 Apr 29  2004 icewm
> >> >-r-xr-xr-x  1 root  wheel  235384 Jan  4  2005 mwm*
> >> >-rwxr-xr-x  1 root  wheel  137384 Jan  4  2005 twm*
> >> >
> >> >Shared libraries and other dynamic extensions add a lot of runtime bloat
> >> >to this.
> >> 
> >> Except that they are, wait for it, shared.
> >
> >That is one of the reasons why they increase bloat.  Their bloat is mostly
> 
> Added GOT/PLT may increase the memory utilization a bit, but it's still
> significantly less than the static version would be.  There will also be

That is not the main reason.  Shared libraries decrease locality.  This
costs 100-200K per active program (thread) in small programs.  Static
linkage of small programs with non-bloated libraries costs less than
that per program and the cost is not repeated for active program.  But
normally the libraries are bloated, so small programs when statically
linked bloat to several times 100-200K, so it takes running several of
them for their memory use to be smaller.

> an imperceptable performance hit from the symbol lookup the first time
> a shared object function is called or shared global variable is accessed.
> Subsequent accesses pay a simple indirection penalty which is obviated
> by the microarchitecture of most modern processors.

The performance hit is perceptible at 3.5 times slower startup and
exit for /bin/sh (starting from a statically linked shell; more for
all-dynamic).  This is with /bin/sh linked to only 3 libraries.  The
slowness is from a combinatation of mapping the libraries, dynamic
linking to them, and later cleaning up larger threads on exit.
All-dynamic gives additional slowness from forking larger threads.
Indirections from the GOT/PLT cost relatively less.

> The key is that the big shared objects (such as libc or the graphics
> libraries, for example) are
> shared amongst _all_ processes, not just instances of the shell.

Part of the bloat is that the data is not shareable and tends to have
poor locality.  This ends up more than doubling the physical memory
use for each extra instance for the version of /bin/sh that I tested.
All versions had relatively small physical memory use --- just 124K
extra for shared linkage.  It only takes 32 variables scattered
across 32 4K pages in shared libraries instead of packed into 1 page
in static data to cost 124K.  This effect is smaller for large programs,

Bruce

[toc] | [prev] | [next] | [standalone]

#9347

bde@besplex.bde.org (Bruce Evans) writes:
>In article <rBOZx.13293$9Q.2428@fx01.iad>,
>Scott Lurndal <slp53@pacbell.net> wrote:

>> Added GOT/PLT may increase the memory utilization a bit, but it's still
>> significantly less than the static version would be.  There will also be
>
>That is not the main reason.  Shared libraries decrease locality.  This

Spatial or temporal?

   Modern processors have  cache and TLB structures to address
   spatial locality issues.

>costs 100-200K per active program (thread) in small programs.  Static

  100-200K what?

If I were running in 1MB of memory, I'd be concerned.   With 16GB,
200k is in the noise, particularly with 64k page sizes (armv8, e.g.).

[toc] | [prev] | [next] | [standalone]

#9251

In comp.misc gareth <no.spam@thank.you.invalid> wrote:
> "Rich" <rich@example.invalid> wrote in message 
> news:n10dgn$uk2$2@dont-email.me...
> > The fvwm2 executable is 862,808 bytes

> Why?

Ambigious question - did you mean: Why is it only that big?  Maybe
because it is free of the bloat that infests so much other software
that tries to be the kitchen sink plus everything else in the house,
all at the same time.

[toc] | [prev] | [next] | [standalone]

#9278

"gareth" <no.spam@thank.you.invalid> writes:

> "Dan Espen" <despen@verizon.net> wrote in message 
>
>> Simple cure.  Run Linux.
>> Don't use any of the highfalutin desktops, just a basic WM.
> 
> Even then, you need several mega bytes of bloat and you do not know
> what is happening underneath.

The original post was to the effect that we all oughta move regularly
to the latest and greatest hardware and software.  If you *do* stick
with one system for years/tech-generations, you gradually learn as
much of what's underneath as your skill level allows.  In effect, you
graually master your tools.  On a personal level your productivity and
effectiveness whould rise.  From a management point of view that
ignores that cognitive element, OTOH, new, allegedly faster and more
sophisticated stuff *must* be better because, well, yew know, it's
*new*.  But "life-long learning" is supposed to be about leaning new
stuff, not learning the same stuff over and over as UI designers and
application programmers add more knobs and chrome and putative
features. 

How do you win a war if, as soon as your pilots lean to fly the
lastest aircraft, you replace them with newer and different aircraft?

-- 
Mike Spencer                  Nova Scotia, Canada

[toc] | [prev] | [next] | [standalone]

#9249

On 2015-10-30, Dan Espen <despen@verizon.net> wrote:
> Simple cure.  Run Linux.
> Don't use any of the highfalutin desktops, just a basic WM.
>

Conceptually, what I do on my Core Duo (2G RAM) isn't much different
than what I did on a Pentium III (128M RAM) over a decade ago (one
exception: video transcoding with Handbrake).

That PIII started feeling slow when Linux systems moved from ISO-8859-X
and friends to Unicode.  Seems like Unicode required more RAM and
therefore a bit of new hardware seemed worthwhile.  Obviously that RAM
is now much appreciated when ripping DVDs to MKVs.  Other than that, I
could probably go back to that Pentium and do most of what I do today
with little alteration.

Windowmaker is my sweetspot for low resource WMs.

[toc] | [prev] | [next] | [standalone]

#9248

On Fri, 30 Oct 2015 13:43:47 +0000, gareth wrote:

> "RS Wood" <rsw@therandymon.com> wrote in message
> news:d9h7ivF1apvU1@mid.individual.net...
>> http://www.theregister.co.uk/2015/10/15/
junk_your_it_now_before_it_drags_you_under/
> 
> 
> Very well said.
> 
> Give me a machine with GHz speed, and astronmical sized hard disk, a
> retina display,
> but otherwise completely lacking in system software, so that I can
> control from the ground up, just as I did with my first experience of a
> PDP-11/20 back in 1971.
> 
> I always revelled in close contact with the machine, and as time has
> progressed, I feel more and more divorced from the computers that I
> love.

 That pretty well echoes my own experience with Microsoft windows from 
win95osr2 and windows 2000 on. To my mind, win2k (classic desktop with 
open each folder in its own window option) was the epitome of 
'connectedness' to the computer hardware which MSFT has been eroding with 
each successive release of its successor OSes.

 In my own case, I solved that problem by simply not "upgrading" from 
win2k. However, that strategy finally failed me with a recent (and long 
overdue) hardware upgrade which forced me to finally upgrade the host OS 
to Linux Mint 17.1 KDE 64 bit (the least retro/klunky option I could see 
available to fit my needs).

 As much as typical windows users might think win7 is now the best option 
out of the current OS offerings from MSFT, that's not the case for me, 
hence my migration to a *nix based solution. Despite the major downsides 
to using a Linux DE, I feel the situation with MSFT's offerings has 
reached such a new low that I'd rather face the disappointments of Linux 
than suffer at the hands of MSFT, especially in view of their latest 
privacy policy enshrined in windows 10 now being pushed onto win7/8 users 
as a "Free Upgrade".

 TBH, until win10, I think MSFT demonstrated a high level of integrity in 
regard of their customers' privacy. Looking at the traffic analysis of 
the "telemetry" that was published recently, it's hard to decide whether 
it was commercial pressure to compete with Google et al in data mining 
their cash cow resource or pressure from the NSA and GCHQ as the 
motivating force behind this 'change of heart'. I suspect it was a 
combination of both.

 The bad news for everyone, expert computer user and mere consumer alike, 
is that the increasing sophistication and complexity of the support chips 
(including CPUs) used in PCs and smart phones and tablets is allowing for 
more sophisticated 'traitorware' to be embedded within the very fabric of 
those electronic devices to the point where the choice of OS will make 
absolutely no difference to the security of your personal and private 
data.

 If anyone is naive enough to think that the use of retro pre-traitorous 
hardware (eg Pentium 2 based PC kit) will make a difference in your 
'visibility' to your nation's security services, think again. Even if you 
stay disconnected from the internet with your retro computing activities, 
the advent of social media means that your immediate friends and 
relatives will be 'ratting you out' to the authorities simply as a side 
effect of their own lack of interest in what others know about 
themselves. That, and the fact that pretty well everything your 
government needs to know about you is now stored and accessed 
electronically means your life remains an open book to those with the 
wherewithal and a motive to take a close look at your profile and more 
recent activities. Trying to 'stay off grid' will likely make you look 
like an interesting target worthy of closer scrutiny simply because of 
your failure to follow the herd mentality.

 Leaving that depressing view of our current situation to one side for 
the time being to get back on topic, it's always struck me, since the 
days of windows 95, that MSFT's mission statement could accurately be 
restated as "Never allow hardware developments to be reflected back to 
the user as improvements in system responsiveness."

 They seem to have seen it as their duty to bloat the OS and apps just 
sufficiently to maintain the user experience typical of 1995 in spite of 
all the potential performance gains in successive generations of PC 
hardware over the past two decades. Quite obviously, the result of the 
MSFT/OEMs cartel's desire to drive the sales of new PC hardware. This 
'cartel effect' is so blatantly obvious as to have been informally given 
the name "Wintel".

 The only hardware improvement that has escaped MSFT's attention being 
the advent of SSD storage devices which have not only improved 
responsiveness but also sidelined the 'rapid ageing' effect of a bunch of 
invented pagefile options designed to accelerate the performance sapping 
effects of runaway FS fragmentation induced by the kakamaimee defaults in 
the single huge partition, 'all your eggs in one basket' setup of the 
typical OEM configured windows PC.

 In short, the answer to the question posed by this thread's subject line 
is a resounding "Hell, yes!" (at least as far as windows afflicted PC 
hardware is concerned).

-- 
Johnny B Good

[toc] | [prev] | [next] | [standalone]

#9303

On Fri, Oct 30, 2015 at 07:53:00PM +0000, Johnny B Good wrote:
> On Fri, 30 Oct 2015 13:43:47 +0000, gareth wrote:
> 
> > "RS Wood" <rsw@therandymon.com> wrote in message
> > news:d9h7ivF1apvU1@mid.individual.net...
> >> http://www.theregister.co.uk/2015/10/15/
> junk_your_it_now_before_it_drags_you_under/
> > 
> > 
> > Very well said.
> > 
> > Give me a machine with GHz speed, and astronmical sized hard disk, a
> > retina display,
> > but otherwise completely lacking in system software, so that I can
> > control from the ground up, just as I did with my first experience of a
> > PDP-11/20 back in 1971.
> > 
> > I always revelled in close contact with the machine, and as time has
> > progressed, I feel more and more divorced from the computers that I
> > love.
> 
>  That pretty well echoes my own experience with Microsoft windows from 
> win95osr2 and windows 2000 on. To my mind, win2k (classic desktop with 
> open each folder in its own window option) was the epitome of 
> 'connectedness' to the computer hardware which MSFT has been eroding with 
> each successive release of its successor OSes.
> 
>  In my own case, I solved that problem by simply not "upgrading" from 
> win2k. However, that strategy finally failed me with a recent (and long 
> overdue) hardware upgrade which forced me to finally upgrade the host OS 
> to Linux Mint 17.1 KDE 64 bit (the least retro/klunky option I could see 
> available to fit my needs).
> 
>  As much as typical windows users might think win7 is now the best option 
> out of the current OS offerings from MSFT, that's not the case for me, 
> hence my migration to a *nix based solution. Despite the major downsides 
> to using a Linux DE, I feel the situation with MSFT's offerings has 
> reached such a new low that I'd rather face the disappointments of Linux 
> than suffer at the hands of MSFT, especially in view of their latest 
> privacy policy enshrined in windows 10 now being pushed onto win7/8 users 
> as a "Free Upgrade".
> 
>  TBH, until win10, I think MSFT demonstrated a high level of integrity in 
> regard of their customers' privacy. Looking at the traffic analysis of 
> the "telemetry" that was published recently, it's hard to decide whether 
> it was commercial pressure to compete with Google et al in data mining 
> their cash cow resource or pressure from the NSA and GCHQ as the 
> motivating force behind this 'change of heart'. I suspect it was a 
> combination of both.
> 
>  The bad news for everyone, expert computer user and mere consumer alike, 
> is that the increasing sophistication and complexity of the support chips 
> (including CPUs) used in PCs and smart phones and tablets is allowing for 
> more sophisticated 'traitorware' to be embedded within the very fabric of 
> those electronic devices to the point where the choice of OS will make 
> absolutely no difference to the security of your personal and private 
> data.
> 
>  If anyone is naive enough to think that the use of retro pre-traitorous 
> hardware (eg Pentium 2 based PC kit) will make a difference in your 
> 'visibility' to your nation's security services, think again. Even if you 
> stay disconnected from the internet with your retro computing activities, 
> the advent of social media means that your immediate friends and 
> relatives will be 'ratting you out' to the authorities simply as a side 
> effect of their own lack of interest in what others know about 
> themselves. That, and the fact that pretty well everything your 
> government needs to know about you is now stored and accessed 
> electronically means your life remains an open book to those with the 
> wherewithal and a motive to take a close look at your profile and more 
> recent activities. Trying to 'stay off grid' will likely make you look 
> like an interesting target worthy of closer scrutiny simply because of 
> your failure to follow the herd mentality.
> 
Ditto.  The average consumer completely fails to understand the
significance of all of their personal information to the designs of
malicious interests.  They naively assume that "the government" or
large corporate interests have no interest in amassing psychometric
dossiers on the entirety of the population, and that they presumed
innocuous nature of their lives and interests provide sufficient
protection from abuses of power.  It may be true that most people will
never have their information abused for criminal or quasi-criminal
purposes, but the mere fact of the existence of their personal
information in clandestine files and databases is a latent threat at
best.

In Canada the RCMP desperately wants to classify environmental
activists in the same category as terrorists.  (Never mind that large
numbers of the RCMP are themselves criminals and terrorists in their
own right.)  Universities are full of spies and recruited informers,
and commercial corporations are increasingly becoming partners with
intelligence agencies where they are not outright front organizations.

The casualty figures may be in the fractional percentage range, but it
is very clear that police and intelligence agencies are making war on
civilian populations in the West and they are leveraging large scale
data mining operations in support.  The assumption that people
commonly make, "I've done nothing wrong; I have nothing to hide", is
incorrect when the motivation for one or another form of attack is to
make someone take actions against their interests for political
purposes.  The industrialization of financial fraud and certain other
forms of property crime is made much easier for those who have access
to great masses of information about the public, and if anyone thinks
the people involved in the intelligence community are above that they
should bloody well think again.

>  Leaving that depressing view of our current situation to one side for 
> the time being to get back on topic, it's always struck me, since the 
> days of windows 95, that MSFT's mission statement could accurately be 
> restated as "Never allow hardware developments to be reflected back to 
> the user as improvements in system responsiveness."
> 
>  They seem to have seen it as their duty to bloat the OS and apps just 
> sufficiently to maintain the user experience typical of 1995 in spite of 
> all the potential performance gains in successive generations of PC 
> hardware over the past two decades. Quite obviously, the result of the 
> MSFT/OEMs cartel's desire to drive the sales of new PC hardware. This 
> 'cartel effect' is so blatantly obvious as to have been informally given 
> the name "Wintel".
> 
>  The only hardware improvement that has escaped MSFT's attention being 
> the advent of SSD storage devices which have not only improved 
> responsiveness but also sidelined the 'rapid ageing' effect of a bunch of 
> invented pagefile options designed to accelerate the performance sapping 
> effects of runaway FS fragmentation induced by the kakamaimee defaults in 
> the single huge partition, 'all your eggs in one basket' setup of the 
> typical OEM configured windows PC.
> 
>  In short, the answer to the question posed by this thread's subject line 
> is a resounding "Hell, yes!" (at least as far as windows afflicted PC 
> hardware is concerned).

It is probably fair to say that in many ways, the art of software
engineering is going backwards.  Rather than a continued refinement of
best practices what we too often see is the accretion of needless
technical debt and a great squandering of finite resources.  To the
last the most critical loss is the resource of productive man years.



--
The practitioners of Freemasonry are the rabid Islamic extremists of
Christiandom.  Heroin makes them invincible and all-knowing, and
therefore superior to mere mortals.  Their primary distinguishing
characteristic is the habit of Orwellian doublethink which permits
them to be uncivilized while faking participation in civil society.
Pound for pound they are umatched for sheer density of stupid.

[toc] | [prev] | [next] | [standalone]

#9281

"gareth" <no.spam@thank.you.invalid> wrote in message 
news:n0vs2g$58m$1@dont-email.me...
> "RS Wood" <rsw@therandymon.com> wrote in message 
> news:d9h7ivF1apvU1@mid.individual.net...
>> http://www.theregister.co.uk/2015/10/15/junk_your_it_now_before_it_drags_you_under/
> Very well said.
> Give me a machine with GHz speed, and astronmical sized hard disk, a 
> retina display,
> but otherwise completely lacking in system software, so that I can control 
> from the
> ground up, just as I did with my first experience of a PDP-11/20 back in 
> 1971.
> I always revelled in close contact with the machine, and as time has 
> progressed, I feel more
> and more divorced from the computers that I love.

In particular, I want absolute control and knowledge of every Internet
communication.

[toc] | [prev] | [next] | [standalone]

#9287

gareth wrote:
> "gareth" <no.spam@thank.you.invalid> wrote in message
> news:n0vs2g$58m$1@dont-email.me...
>> "RS Wood" <rsw@therandymon.com> wrote in message
>> news:d9h7ivF1apvU1@mid.individual.net...
>>> http://www.theregister.co.uk/2015/10/15/junk_your_it_now_before_it_drags_y
ou_under/
>> Very well said.
>> Give me a machine with GHz speed, and astronmical sized hard disk, a
>> retina display,
>> but otherwise completely lacking in system software, so that I can control
>> from the
>> ground up, just as I did with my first experience of a PDP-11/20 back in
>> 1971.
>> I always revelled in close contact with the machine, and as time has
>> progressed, I feel more
>> and more divorced from the computers that I love.
>
> In particular, I want absolute control and knowledge of every Internet
> communication.

Get a 4800 baud external modem.

/BAH

[toc] | [prev] | [next] | [standalone]

#9308

"jmfbahciv" <See.above@aol.com> wrote in message 
news:PM0005237ACBAE328A@aca42838.ipt.aol.com...
> gareth wrote:
>> "gareth" <no.spam@thank.you.invalid> wrote in message
>> news:n0vs2g$58m$1@dont-email.me...
>>> "RS Wood" <rsw@therandymon.com> wrote in message
>>> news:d9h7ivF1apvU1@mid.individual.net...
>>>> http://www.theregister.co.uk/2015/10/15/junk_your_it_now_before_it_drags_y
> ou_under/
>>> Very well said.
>>> Give me a machine with GHz speed, and astronmical sized hard disk, a
>>> retina display,
>>> but otherwise completely lacking in system software, so that I can 
>>> control
>>> from the
>>> ground up, just as I did with my first experience of a PDP-11/20 back in
>>> 1971.
>>> I always revelled in close contact with the machine, and as time has
>>> progressed, I feel more
>>> and more divorced from the computers that I love.
>>
>> In particular, I want absolute control and knowledge of every Internet
>> communication.
>
> Get a 4800 baud external modem.

Stupid way to get that. 

[toc] | [prev] | [next] | [standalone]

#9315

In article <d9ntivFmmv4U1@mid.individual.net>,
Rod Speed <rod.speed.aaa@gmail.com> wrote:
>
>
>"jmfbahciv" <See.above@aol.com> wrote in message 
>news:PM0005237ACBAE328A@aca42838.ipt.aol.com...
>> gareth wrote:
>>> "gareth" <no.spam@thank.you.invalid> wrote in message
>>> news:n0vs2g$58m$1@dont-email.me...
>>>> "RS Wood" <rsw@therandymon.com> wrote in message
>>>> news:d9h7ivF1apvU1@mid.individual.net...
>>>>> http://www.theregister.co.uk/2015/10/15/junk_your_it_now_before_it_drags_y
>> ou_under/
>>>> Very well said.
>>>> Give me a machine with GHz speed, and astronmical sized hard disk, a
>>>> retina display,
>>>> but otherwise completely lacking in system software, so that I can 
>>>> control
>>>> from the
>>>> ground up, just as I did with my first experience of a PDP-11/20 back in
>>>> 1971.
>>>> I always revelled in close contact with the machine, and as time has
>>>> progressed, I feel more
>>>> and more divorced from the computers that I love.
>>>
>>> In particular, I want absolute control and knowledge of every Internet
>>> communication.
>>
>> Get a 4800 baud external modem.
>
>Stupid way to get that. 

Go get an OpenBSD (or FreeBSD) firewall, using pf. And use wireshark
and/or tcpdump extensively (on the firewall) to see what hosts are
indeed contacted from your device. And use pf to block, log or remap
the calls you see as suspicious. A squid on a gateway host can also
help a lot, there you can remap on the urls; you don't need to get the
hosts. 

I still trust those, even if I don't trust Linux any longer due to
the insane bloat of libraries. The BSDs are still trustworthy because 
of the "thousand eyes" effect, IMNSHO, on a rather smaller and more
organised code base.

I do this with my home network, and have travelling hosts on a separate
network, both physical and wifi; so they cannot directly contaminate
my internal hosts.

(Yes, I do write this on a travelling host).

This way you can see a lot more than on a 4800 baud modem, and you
can filter and block whatever you don't like. I even use a squid
plugin to analyse the downloads, and drop the ones that seem unsafe.
This drops around 90% of the flash animated advertisements; and I
get chastised by lots of web sites for running an ad blocker. No, 
I don't block them because they are ads, but because they have the
signs of compromised code.

I did a consulting job for a large newspaper about what the ads
they linked to contained. That was a huge wakeup call for both me
and them. Now they run all ads served inhouse, and do a solid wash
of the files being uploaded, and do flash strips and jpeg compression
to make them safe and also a LOT smaller. 

They save around 1 gigabit just from this, from a top load of 6-7.

Barb: Yes, there was a time when a 4800 baud modem with a protocol
analyser was state of the art. That was when there were still more
PDP10s to be announced, the VAX was just out, and Prime had just
come out with workable 50-series machines.

-- mrr

[toc] | [prev] | [next] | [standalone]

#9353

In alt.folklore.computers gareth <no.spam@thank.you.invalid> wrote:
> "RS Wood" <rsw@therandymon.com> wrote in message 
> news:d9h7ivF1apvU1@mid.individual.net...
> > http://www.theregister.co.uk/2015/10/15/junk_your_it_now_before_it_drags_you_under/
> 
> 
> Very well said.
> 
> Give me a machine with GHz speed, and astronmical sized hard disk, a retina 
> display,
> but otherwise completely lacking in system software, so that I can control 
> from the
> ground up, just as I did with my first experience of a PDP-11/20 back in 
> 1971.
> 
> I always revelled in close contact with the machine, and as time has 
> progressed, I feel more
> and more divorced from the computers that I love.
> 

Plying devils advocate: much of software bloat is forced by
hardware.  A simple example is hard dirives.  Random access
time to hard drives improved maybe 5 times compared to 196x
era.  At the same time sequential read speed improved about
1000 times and capacity about 10^5 times.  Note that simple
operation "read the whole drive" now takes _much more_ time
than in the past (capacity increased faster than read speed).
Basically, without disk cache and large block transfers you
will get speed only slightly better than PDP-11/20.

If you look at USB you will see that the specification
actually forces rather complicated implementation.  Some
folks are very critial about USB.  But face it: USB can
do several transfers to different devices without any
CPU activity during transfers -- you set up transactions,
let USB do its work and then collect results.  And USB
support wide range of peripherials, with ability to hot
plug and autoconfigure devices.  Would  you like to
run sysgen every time before you insert a USB gadget?
Or have a setup were USB socket number 1 is dedicated
to keyboard, number 2 to mouse, number 3 to your specific
pendrive, etc. and trying to insert different device or
inserting device in wrong slot does not work.

In MSDOS era hardware drivers could be simple, because
driver frequently would take over machine for duration
of the whole operation.  More advanced version would
work with interrupts.  But taking an interrupt (or
several) for transfer would lead to very bad speed on
modern machines -- modern hard drives and network cards
can transfer several dectors (or packets) per interupt.

On different level, users now demand computers working
in their native language.  So you either have each
country developinig their own national software (clearly
unfeasible) or some configurables codepages (workable
but messy) or Unicode.  Now, Unicode has more than 10^6
potential characters and more than 10^5 assigned, with
useful charactes spread out in several blocks.  The
old school text software would use tables with 128
entries (one for each ASCII code) to make decision.
Even several tables fits in few kB of memory.  Naively
using the same approach with Unicode leads to tens
of MB memory use.  Worse, some tables may need to be
initialized at runtime and than it leads to huge
slowdown.  So someting smarter is meeded.  But this
leads to much more complicated code which is still
slower than ASCII version.  And this is just at very
basic level.  On top of that you need to handle
varying text directions, complicated sorting
rules, etc.  Basically you need large tables to
describe rules of various languages and special
libraries.

An extra anecdote: some time ago I looked why Linux kernel
after boot took that much memory (more than 10 MB).  Then
I realised that to manage memory Linux has tables with
entries per each pago of physical memory.  Compared to
memory present in the machine the tables are tiny.  And
they were "always" present -- simply on small machines
kernel code took much larger percentage of available
memory and memory usage of tables did not stand out.
In old days operating sytem would take sizable part
of available memory, frequently more than half.  Modern
Linux kernel takes only tiny part of memory, so
relative overhead is quite small.  

-- 
                              Waldek Hebisch

[toc] | [prev] | [next] | [standalone]

#9354

In article <n1bb5v$4u$1@z-news.wcss.wroc.pl>,
Waldek Hebisch  <hebisch@math.uni.wroc.pl> wrote:
>In alt.folklore.computers gareth <no.spam@thank.you.invalid> wrote:
>> "RS Wood" <rsw@therandymon.com> wrote in message 
>> news:d9h7ivF1apvU1@mid.individual.net...
>> > http://www.theregister.co.uk/2015/10/15/junk_your_it_now_before_it_drags_you_under/
>> 
>> 
>> Very well said.
>> 
>> Give me a machine with GHz speed, and astronmical sized hard disk, a retina 
>> display,
>> but otherwise completely lacking in system software, so that I can control 
>> from the
>> ground up, just as I did with my first experience of a PDP-11/20 back in 
>> 1971.
>> 
>> I always revelled in close contact with the machine, and as time has 
>> progressed, I feel more
>> and more divorced from the computers that I love.

With a BSD you get a LOT more close to the hardware. There
is a distinct "look and feel" of snappy response with all of them
when run on modern hardware.

>Plying devils advocate: much of software bloat is forced by
>hardware.  A simple example is hard dirives.  Random access
>time to hard drives improved maybe 5 times compared to 196x
>era.  At the same time sequential read speed improved about
>1000 times and capacity about 10^5 times.  Note that simple
>operation "read the whole drive" now takes _much more_ time
>than in the past (capacity increased faster than read speed).
>Basically, without disk cache and large block transfers you
>will get speed only slightly better than PDP-11/20.

[snip]

>old school text software would use tables with 128
>entries (one for each ASCII code) to make decision.
>Even several tables fits in few kB of memory.  Naively
>using the same approach with Unicode leads to tens
>of MB memory use.  Worse, some tables may need to be
>initialized at runtime and than it leads to huge
>slowdown.  So someting smarter is meeded.  But this
>leads to much more complicated code which is still
>slower than ASCII version.  And this is just at very
>basic level.  On top of that you need to handle
>varying text directions, complicated sorting
>rules, etc.  Basically you need large tables to
>describe rules of various languages and special
>libraries.

Still, UTF-8 caches the "codepage", and takes less than
twice the space of plain old ASCII; even when doing some
language with lots of character modifiers, like thai or
arabic.

>An extra anecdote: some time ago I looked why Linux kernel
>after boot took that much memory (more than 10 MB).  Then
>I realised that to manage memory Linux has tables with
>entries per each pago of physical memory.  Compared to
>memory present in the machine the tables are tiny.  And
>they were "always" present -- simply on small machines
>kernel code took much larger percentage of available
>memory and memory usage of tables did not stand out.
>In old days operating sytem would take sizable part
>of available memory, frequently more than half.  Modern
>Linux kernel takes only tiny part of memory, so
>relative overhead is quite small.  

Indeed. On the machine I write this on, an early ARM7 with
2G mamory Linux including all drivers and tables take 22.5 
megabytes, and the tables take around half of that. So
Linux with drivers take somewhat more then 10. Linux reports
3679 kb in various lockable tables, and each pid takes 1152
bytes. 256k is for a DMA pool. Nothing particularly wasteful
here. With hardware around 2000 times faster, 250+ times
more disk and memory and support for around 17k various
peripherals and I/O speeds >> 100x what we used ca 1980, but
the code size of the kernel has gone up around 10-20x. I
call that very good.

Still only twice as big as VMS got to be while it was
still on VAXen, and still within an order of magnitude
from Tops20; which has around 5% as much space used by
drivers. (How many drivers are in Tops20 all in all? 100
maybe?)

-- mrr

[toc] | [prev] | [next] | [standalone]

#9377

Morten Reistad wrote:
> In article <n1bb5v$4u$1@z-news.wcss.wroc.pl>,

> Still only twice as big as VMS got to be while it was
> still on VAXen, and still within an order of magnitude
> from Tops20; which has around 5% as much space used by
> drivers. (How many drivers are in Tops20 all in all? 100
> maybe?)

A lot, lot less.  One LPT; 4 or 5 disks; one terminal
2 magtapes; and then the stuff needed if one used DECnet.

/BAH

[toc] | [prev] | [next] | [standalone]

#9355

On 03/11/2015 22:06, Waldek Hebisch wrote:
> In alt.folklore.computers gareth <no.spam@thank.you.invalid> wrote:

>> I always revelled in close contact with the machine, and as time has
>> progressed, I feel more
>> and more divorced from the computers that I love.
>>
>
> Plying devils advocate: much of software bloat is forced by
> hardware.  A simple example is hard dirives.  Random access
> time to hard drives improved maybe 5 times compared to 196x
> era.  At the same time sequential read speed improved about
> 1000 times and capacity about 10^5 times.  Note that simple
> operation "read the whole drive" now takes _much more_ time
> than in the past (capacity increased faster than read speed).
> Basically, without disk cache and large block transfers you
> will get speed only slightly better than PDP-11/20.

I don't understand why increasing capacity of memory or storage, or 
running at faster clock speed, by itself increases complexity. You might 
need bigger pointers to address memory, but at worst that would double 
the size of the data used by a program.

> If you look at USB you will see that the specification
> actually forces rather complicated implementation.

I don't know what's involved with driving USB, but I don't see that it's 
so complex that it would need hundreds of thousands of lines of code.

30 years ago I was printing graphics on dot matrix printers. The drivers 
were a few KB, with a few dozen lines of difference between Epson and 
HP. Now, a driver download just for HP might be 40MB** (filling four 
entire hard drives 30 years ago). (And, having done it, I can tell you 
that generating a raster image for a laser printer isn't any harder than 
doing it for an dot matrix printer. There's just more data.)

(** And that particular download didn't solve the problem I was having! 
One subsystem of the computer had no idea what another subsystem was doing.)

> Now, Unicode has more than 10^6
> potential characters and more than 10^5 assigned, with
> useful charactes spread out in several blocks.  The
> old school text software would use tables with 128
> entries (one for each ASCII code) to make decision.
> Even several tables fits in few kB of memory.  Naively
> using the same approach with Unicode leads to tens
> of MB memory use.  Worse, some tables may need to be
> initialized at runtime and than it leads to huge
> slowdown.  So someting smarter is meeded.  But this
> leads to much more complicated code which is still
> slower than ASCII version.  And this is just at very
> basic level.  On top of that you need to handle
> varying text directions, complicated sorting
> rules, etc.  Basically you need large tables to
> describe rules of various languages and special
> libraries.

Unicode means using 2 or 4 bytes per character instead of 1, that's all. 
And storage schemes such as UTF8 mean that text files needn't take up 
that much more space. As for those that tables indexed by character: you 
simply wouldn't have a table with 64K or 1M entries; you'd use another 
approach (and probably for most purposes, special characters will be 
part of the ASCII subset).

Supporting Unicode with a big selection of fonts means a lot more data, 
but not necessarily that much more complexity (not of the order of 100 
or 1000 anyway). You would also expect such software to be optimised for 
the default language.

(Also, you simply don't use tables -- for a parser etc -- that span 64K 
or 1M characters instead of 256. Programming languages for example still 
tend to use English.)

> An extra anecdote: some time ago I looked why Linux kernel
> after boot took that much memory (more than 10 MB).  Then
> I realised that to manage memory Linux has tables with
> entries per each pago of physical memory.  Compared to
> memory present in the machine the tables are tiny.  And
> they were "always" present -- simply on small machines
> kernel code took much larger percentage of available
> memory and memory usage of tables did not stand out.
> In old days operating sytem would take sizable part
> of available memory, frequently more than half.  Modern
> Linux kernel takes only tiny part of memory, so
> relative overhead is quite small.

On CP/M it might have been 8KB out of 64KB. One machine I used also used 
to take 1.5 seconds to boot (from floppy!), before I could start doing 
useful work. Most machines now take far longer when starting from 'cold' 
(an Android tablet I have takes *two minutes*. I would really love to 
know what on earth it's doing!).

-- 
Bartc

[toc] | [prev] | [next] | [standalone]

#9360

In alt.folklore.computers BartC <bc@freeuk.com> wrote:
> On 03/11/2015 22:06, Waldek Hebisch wrote:
> > In alt.folklore.computers gareth <no.spam@thank.you.invalid> wrote:
> 
> >> I always revelled in close contact with the machine, and as time has
> >> progressed, I feel more
> >> and more divorced from the computers that I love.
> >>
> >
> > Plying devils advocate: much of software bloat is forced by
> > hardware.  A simple example is hard dirives.  Random access
> > time to hard drives improved maybe 5 times compared to 196x
> > era.  At the same time sequential read speed improved about
> > 1000 times and capacity about 10^5 times.  Note that simple
> > operation "read the whole drive" now takes _much more_ time
> > than in the past (capacity increased faster than read speed).
> > Basically, without disk cache and large block transfers you
> > will get speed only slightly better than PDP-11/20.
> 
> I don't understand why increasing capacity of memory or storage, or 
> running at faster clock speed, by itself increases complexity. You might 
> need bigger pointers to address memory, but at worst that would double 
> the size of the data used by a program.

If you accept fact that your new high-powered maching will
be only slightly faster then 40 years old one providing the
the same features as the old one, then there is no need
for more complexity.  But at comparable featurs modern
machine may do much more work: typical Linux desktop is
able to handle more than 1000 telent session (for 1000
users).  PDP-11/20 probably would be overloaded at 10
(and lack features like command line history present
in modern command line).  Without disc cache machine
with single disc drive would get overloaded probably
below 40 users.  You need complexity to fully
hardware potential.

You sound like hardware were working as in the post,
only faster and bigger, basicaly just scaling up.
Even if it ere simple scaling, you stil may need smarter
algorithms, for example linearly scannig all processes on
each clock tick is fine if you have 10 processes, but becames
problematic when there are thousends of them.  But hardware
progress is anything like that: DRAM latency is about 50ns
which is only slightly better than 70ns in 1990, and only
few time better than 250-350ns from late seventies.  So,
current machines are fast only due to caches.  And with
caches something as simple as matrix multiplication sudenly
needs smart algorithm: you can use just three simple loops
or library lile ATLAS.  Only problem is that simple method
is something like 500 time slower for big matrices.  Similarly
on system level: cache effectively becomes part of context
that is stored and restored on interrupts.  So, when
uniterrupted modern processor is quite fast, but time
lost due to an interrupt may be bigger than on simpler
(and otherwise slower) processor.  Consequently, you need
to minimize number of interrupts and maximize work per interrupt.  

> > If you look at USB you will see that the specification
> > actually forces rather complicated implementation.
> 
> I don't know what's involved with driving USB, but I don't see that it's 
> so complex that it would need hundreds of thousands of lines of code.
> 
> 30 years ago I was printing graphics on dot matrix printers. The drivers 
> were a few KB, with a few dozen lines of difference between Epson and 
> HP. Now, a driver download just for HP might be 40MB** (filling four 
> entire hard drives 30 years ago). (And, having done it, I can tell you 
> that generating a raster image for a laser printer isn't any harder than 
> doing it for an dot matrix printer. There's just more data.)

Compare resolutions and available fonts: there are a lot more data
needed to get features of modern printer.  OTOH, this also
depends on printer: for my printer I actually use no special
driver (just the genric USB printer support build into Linux
kernel).  I do not have access to some features, but printer
is doing what I need os this is OK.  And, no, data sent to
my printer usually is not raster data: it is Postscript
printer and most data is in vector form (in software I
choose "generic Postscript printer").

> > Now, Unicode has more than 10^6
> > potential characters and more than 10^5 assigned, with
> > useful charactes spread out in several blocks.  The
> > old school text software would use tables with 128
> > entries (one for each ASCII code) to make decision.
> > Even several tables fits in few kB of memory.  Naively
> > using the same approach with Unicode leads to tens
> > of MB memory use.  Worse, some tables may need to be
> > initialized at runtime and than it leads to huge
> > slowdown.  So someting smarter is meeded.  But this
> > leads to much more complicated code which is still
> > slower than ASCII version.  And this is just at very
> > basic level.  On top of that you need to handle
> > varying text directions, complicated sorting
> > rules, etc.  Basically you need large tables to
> > describe rules of various languages and special
> > libraries.
> 
> Unicode means using 2 or 4 bytes per character instead of 1, that's all. 
> And storage schemes such as UTF8 mean that text files needn't take up 
> that much more space. As for those that tables indexed by character: you 
> simply wouldn't have a table with 64K or 1M entries; you'd use another 
> approach (and probably for most purposes, special characters will be 
> part of the ASCII subset).

Exactly, you need another approach, like hash tables with entries
only for relevant characters, or lists or ranges.

> Supporting Unicode with a big selection of fonts means a lot more data, 
> but not necessarily that much more complexity (not of the order of 100 
> or 1000 anyway). You would also expect such software to be optimised for 
> the default language.

In ASCII era you would have a table with charater properties
which can take 128 bytes (for 8 binary properties).  Together
with accessor functions less than 1kb.  Corresponding
library for Unicode when I looked at it were 2.7 MB.  One
reason was that there is a lot of Unicode charaters.  But
there is also more relevant properties.  Given all functionality
the 2.7 MB actually did not look so large.  Note: I am not
talking about fonts, just charater properties.  For example,
in ASCII in "fixed" font all characters are of the same
width, but Unicode has notion of double with charaters.
ASCII is left to right, but Unicode has several properties
to allow mixing left to right with right to left.  There
is so called bidi algorithm which specifies how logical
text should be reordered for display.

Let me add: it is easy to criticise modern systems.  I hate
bloat at least as other folks here.  But _part_ of complexity
is due to genuinely useful features (among them need to drive
modern hardware at adequate performance level).  So comparisons
to PDP11, MVS, MSDOS or even early Linux miss the point:
even hardcore hackers want enough features to get insanely
complex system.  Sure, a lot of new features look pointless.
And at least theoretically it is possible to get much
smaller and still very useful system.  But deciding
between features (which one should be supported) it
pretty hard job.  And if useful feature gets rejected in
its right place, then it reapears in other places
as kludges or workarounds leading to bloat which we
tried to avoid.  So it is not clear if we can do
better.

Brooks wrote: "the systems we build are at the border
of what we can do" (this is from memory, he almost shurely
used different words but the sense was as above).
Assuming that complex system can be built, increasing
complexity we can gain useful features and solve various
problems.  So there is strong pressure to more complex
systems.  And general public have little understanding
of complexity, but can easily spot differences in features.
So market prefers complex solutions and ensures that
system are as complex as we can make them without evident
breakdown.

-- 
                              Waldek Hebisch

[toc] | [prev] | [next] | [standalone]

#9364

In article <n1c1us$c9i$1@z-news.wcss.wroc.pl>,
Waldek Hebisch  <hebisch@math.uni.wroc.pl> wrote:
>In alt.folklore.computers BartC <bc@freeuk.com> wrote:
>> On 03/11/2015 22:06, Waldek Hebisch wrote:
>> > In alt.folklore.computers gareth <no.spam@thank.you.invalid> wrote:
>> 
>> >> I always revelled in close contact with the machine, and as time has
>> >> progressed, I feel more
>> >> and more divorced from the computers that I love.
>> >>
>> >
>> > Plying devils advocate: much of software bloat is forced by
>> > hardware.  A simple example is hard dirives.  Random access
>> > time to hard drives improved maybe 5 times compared to 196x
>> > era.  At the same time sequential read speed improved about
>> > 1000 times and capacity about 10^5 times.  Note that simple
>> > operation "read the whole drive" now takes _much more_ time
>> > than in the past (capacity increased faster than read speed).
>> > Basically, without disk cache and large block transfers you
>> > will get speed only slightly better than PDP-11/20.
>> 
>> I don't understand why increasing capacity of memory or storage, or 
>> running at faster clock speed, by itself increases complexity. You might 
>> need bigger pointers to address memory, but at worst that would double 
>> the size of the data used by a program.

I adressed the Linux kernel size in a post yesterday. It
has grown to about ten times the size of what tops20 took in
1984. Half of it is tables and buffers, nearly 4mb of various
core tables, a few megabytes of buffers for network, dma, 
terminals, disks etc, and around 2% of memory for the 4k 
page tables. It could grow to more than 50mb if all the
loadable drivers are in; but that is unfair, no such machine
exist that would need that.

In a way, Linux is the continuation of Tops20. When you start
to read the kernel code you will understand. It just hides
under the "unix" layer.

>If you accept fact that your new high-powered maching will
>be only slightly faster then 40 years old one providing the
>the same features as the old one, then there is no need
>for more complexity.  But at comparable featurs modern
>machine may do much more work: typical Linux desktop is
>able to handle more than 1000 telent session (for 1000
>users).  PDP-11/20 probably would be overloaded at 10
>(and lack features like command line history present
>in modern command line).  Without disc cache machine
>with single disc drive would get overloaded probably
>below 40 users.  You need complexity to fully
>hardware potential.

But,. it still takes about one minute to do a full build
of emacs from sources, this has been fairly constant. On our
2-processor mips3k machine in ca 1990 it took one minute.
On our 8-way brand new r6k one almost a decade later, same
time. And on my new 8way 2.4Ghz arm7 with spiffy SSD, still
the same time. Emacs has grown. A lot. 

But it loads in two-digit milliseconds now, almost too fast
to see. (unless it fires up over X, which still take a few
seconds). I remember when it took around 40 seconds.

[snip]
>Brooks wrote: "the systems we build are at the border
>of what we can do" (this is from memory, he almost shurely
>used different words but the sense was as above).
>Assuming that complex system can be built, increasing
>complexity we can gain useful features and solve various
>problems.  So there is strong pressure to more complex
>systems.  And general public have little understanding
>of complexity, but can easily spot differences in features.
>So market prefers complex solutions and ensures that
>system are as complex as we can make them without evident
>breakdown.

In a way, the whole western civilisation was struggling 
hard with coping with complexity until they managed to
get it programmed sometime in the 1980s or thereabouts.

Remember all the tables we used in advanced school in the
60s and 70s? Log tables, sine tables, stress tables, 
interest tables and whatnot? And the constant errors in 
payrolls, long delays in interest payments in banks, 
insane times used by the people computing stress levels
on machinery and buildings?

Now these are all computerised, and the calulations take
less than a second, in general. (if we still do it the
way they did it back then. But we DO use more, sometimes
way more advanced approaces now, because we can.)

So, the early computers were sorely needed when they
arrived, and were pushed to their limits. It was the
"Risc Revolution" ca 1987 that finally gave us a leeway
in computing time, not having to work nights any longer.

That was when the computers had caught up with the pent-up
demand in society for handling complexity. After that
it has been inside-out, the computers have been driving
the development.

-- mrr

[toc] | [prev] | [next] | [standalone]

#9367

On Wed, 4 Nov 2015 10:24:15 +0100
Morten Reistad <first@last.name.invalid> wrote:

> And on my new 8way 2.4Ghz arm7 with spiffy SSD

	That sounds rather nice - what is it ? I can't seem to find
anything near those specs

-- 
Steve O'Hara-Smith                          |   Directable Mirror Arrays
C:>WIN                                      | A better way to focus the sun
The computer obeys and wins.                |    licences available see
You lose and Bill collects.                 |    http://www.sohara.org/

[toc] | [prev] | [next] | [standalone]

#9374

In article <20151104095533.c7ee90dff7117a3a75e2ebea@eircom.net>,
Ahem A Rivet's Shot  <steveo@eircom.net> wrote:
>On Wed, 4 Nov 2015 10:24:15 +0100
>Morten Reistad <first@last.name.invalid> wrote:
>
>> And on my new 8way 2.4Ghz arm7 with spiffy SSD
>
>	That sounds rather nice - what is it ? I can't seem to find
>anything near those specs

It is an overclocked, aggressively cooled (-60C) hardkernel xu4, 
runs normally on 2GHz.

http://www.hardkernel.com/main/products/prdt_info.php?g_code=G143452239825

It only uses around 8 watts while running the large builds, and around 1.5
in normal editing, so it is not a very difficult one to cool. I use an ancient
lab-freezer of ~12W effect to cool it and a rpi. Total power consumption for
the freezer and both of them are still << 50W.

-- mrr

[toc] | [prev] | [next] | [standalone]

#9378

On 04/11/2015 04:35, Waldek Hebisch wrote:
> In alt.folklore.computers BartC <bc@freeuk.com> wrote:

>> I don't understand why increasing capacity of memory or storage, or
>> running at faster clock speed, by itself increases complexity.

> If you accept fact that your new high-powered maching will
> be only slightly faster then 40 years old one providing the
> the same features as the old one, then there is no need
> for more complexity.

When you speed up a processor clock by 1000 times, why shouldn't the 
same program run faster?

Modern software is more sophisticated, but it's mostly the extra memory 
and storage that makes it possible. Does it really need a 1000x faster 
machine just to run at the same speed as the old one?

(In 1981 I wrote a little assembler for Z80, running on a 2.5MHz Z80 
(where each 8-bit instruction took between 4 and 24 clock ticks). It 
processed source code at 1700 lines per second.

I recently tested NASM, running on a 3GHz 64-bit x86 processor (with 
multiple 64-bit operations per clock), and it managed 23000 lines per 
second, only 13 times as fast!

There is some very fast software around, but equally a lot of it is 
inexplicably slow.)

> caches something as simple as matrix multiplication sudenly
> needs smart algorithm:

Smart algorithms don't necessarily mean huge bloated code. If something 
needs a million lines of code, then it's not an algorithm! (1M lines is 
about 15000 pages; even the proof of Fermat's Last Theorem was only 
about 100.)

(I use my own language for programs for which other people use C. And I 
run my own compiler for that. That compiler is about 0.3MB.

It does pretty much the same job as, for example gcc, for which a 
typical installation might be 300MB. And even that is small compared 
with Visual C, for which you're looking at a 5000-11000 MB download.

See, it is still possible to keep things small and manageable. But 
programs don't need to be as tiny as mine, they just don't really need 
to be 1000 to 10000 times bigger.)

  you can use just three simple loops
> or library lile ATLAS.  Only problem is that simple method
> is something like 500 time slower for big matrices.

The sort of complexity I'm concerned with is not really about such 
low-level algorithms. It's about systems getting too big enough that no 
one really knows what is going on.

People build on top of other systems which are built on top of others, 
and eventually you've got a horrible bloated mess. And a slow one!

(I know I will be castigated for using Windows in this apparently 
Unix-dominated group, but here goes. Until recently, every morning on my 
PC, a program called SVCHOST would take over my machine. It occupied one 
core of two, but still, everything else practically ground a halt.

It carried on for some half an hour. If stopped, it would start again, 
or there would be problems with connectivity later. It wasn't a virus 
either!

It turned out to be something to do with Windows updates, but it never 
reported any actual updates, it never actually /did/ anything. It just 
made the machine - my machine that I want to work on - impossible to 
use. Other people had the same problems.

It was solved by disabling something to do with the updates. The 
question remains, why would MS allow something like this to run that 
would effectively cripple customers' machines? What exactly was it doing 
that required so much CPU power? Does anyone know for sure? I doubt it!)

>> Unicode means using 2 or 4 bytes per character instead of 1, that's all.
>> And storage schemes such as UTF8 mean that text files needn't take up
>> that much more space. As for those that tables indexed by character: you
>> simply wouldn't have a table with 64K or 1M entries; you'd use another
>> approach (and probably for most purposes, special characters will be
>> part of the ASCII subset).
>
> Exactly, you need another approach, like hash tables with entries
> only for relevant characters, or lists or ranges.
>
>> Supporting Unicode with a big selection of fonts means a lot more data,
>> but not necessarily that much more complexity (not of the order of 100
>> or 1000 anyway). You would also expect such software to be optimised for
>> the default language.
>
> In ASCII era you would have a table with charater properties
> which can take 128 bytes (for 8 binary properties).  Together
> with accessor functions less than 1kb.  Corresponding
> library for Unicode when I looked at it were 2.7 MB.

That's just data. You flash an image onto the screen; that's 1MB or 2MB; 
it's nothing, because the capacity is there.

What's exasperating (this is again Windows), is you right-click 
something (a file, icon, whatever), and sometimes you can count slowly 
to ten before it even shows a pop-up list of options. You're not asking 
for some complicated calculations, you just want it to show 'Properties' 
so that you can click on it!

So what was it spending ten seconds doing? The problem is there's 
apparently no-one whose job it is to find out. They only know how to add 
complexity: buy more RAM, a faster machine, download add-on software to 
'clean' your system, or download yet another upgrade!

-- 
Bartc

[toc] | [prev] | [next] | [standalone]

Page 3 of 5 — ← Prev page 1 2 [3] 4 5  Next page →

Back to top | Article view | comp.misc

csiph-web