[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Faster Indy?



> 
> On Wed, 21 Jul 1999, Donovan wrote:
> 
> >
> > However, given the fact that it probably isn't that hard to do ( just
> > modify RPM_OPT_FLAGS ) there may be merit in doing this. Does optimising
> > for these processors make the binaries unusable or just suboptimal on
> > others ? if the latter, perhaps it's a good idea.
> 

Problem is time and bandwidth.  Compiling with different flags than
those used by RedHat means uploading 600 Megs and it happens my
bandwidth is 10 megs/hour and isn't cheap.  I cannot do it myself.


> According to both the Mandrake and Stampede people there may be a slight
> performance hit on non-Pentium processors (i386/i486 and comp), but the
> binaries are usable. On Pentium and above (Pro,II,III) there is a
> performance gain of 10 - 30% (using the PGCC compiler of coarse).
> Stampede removes all debugging code from the binaries to reduce size
> and gain speed. I'm not sure about Mandrake. Mandrake 6.0 is (claims) to
> be 99.99% compatible w/Red Hat 6.0
> 
> > 
> > It would be interesting to see if there really is any performance benefit
> > ( besides the placebo effect ) Of course, even placebo effects keep users
> > happy (-; 
> 

I don't like what Mandrake is doing now.  First of all they use
hdparms for accelerating the HD.  Problem is that this is to play
Russian roulette with the user's data.  It can be _very_ useful with
some disks but there aren't parms syou are sure they will work on
every disk and every motherboard under thge sun.  I corrupted a disk
with it, and a few months later a guy writing for a Ziff Davis
magazine corrupted his disk after using the "safe" parms I recommended
him.  Now this guy is one of the three Mandrake bosses and is having
Mandrake using hdparms on the user's disks.  About the settings
Mandrake is using I made a read only test on my three disks and my
best one put itself into non DMA mode after losing an interrupt.  This
slowed it by a factor of three and increased the CPU consumption by a
factor of five thus making it nearly unusable on a multitasking
system.  I didn't dare to make read-write tests using Madrake's
parms.  So hdparms is something you play with while using a single
filesystem on an expendable partition until you find the right parms
NOT something you use on a distribution.

Other problems I found: they are distributing true type fonts who
cannot be distributed commercially, they include two mail agents
despite the fact they dont't have resolved the problem of conflicting
packages during install and one of those agents is postfix whose
present license allows IBM to ask you to stop using it at IBM's entire
discretion.

Allowing such things in a distribution is IMHO grossly unethical.  Of
course you are free to think my opinion is due to jealousy. :-)

> IMO I think that the optimizations are slightly faster on my Pentium
> 200 MMX. Stampede has a benchmark listed on it's web site at
> http://www.stampede.org/faq.php3 if anyone is interested.
> 

A better idea would be to include the MMX library in case there is one
(anybody wants to look for it?).  It seems Sane could use it and this
would accelerate considerably.  It would then be child's play to make
a shell wrapper for choosing the MMX orr the noraml version.

> I've been doing some R & D on some other distro's and awhile back I
> downloaded and installed Mandrake 6.0 and have been using it. I have
> tried many distro's in the past several months. I'm trying to get the
> feel for some of the major distro's so I can form my own opinions (on
> strengths and weaknesses) and gain better insight on installations etc.
> I've got my eye on the Loather project (~8
> 

Yes I have been thinking in it.  The fact Windows does handle so well
hardware recognition is due to no magic but to the fact the driver is
in the box with your hardware.

So IMHO the best aproach is not writing drivers or having a project
for hardware recognition (useful only with ISA cards) but quite simply
use the power of the almighty buck.  First of all when you don't have
adriver for some hardware tell the manufacturer how unhappy you are
and how this will influence your future buyings.  Second lobby Ziff
Davis for some comparative tests of hardware under Linux: some
customers will find that such card is much slower than such other due
to a a poor driver so they will refrain to buy them and this will push
manufacturers to a more cooperative attitude and even to write the
drivers themselves.  This is the way to have hardware recognition and
drivers for the latest and greatest cards instead of waiting for
months until a Linuxer buys one and uses it for writing a driver.

> So far I think the approach Indy has taken is the best so far. And I'm
> not just saying that to get attention! IMHO (and others!) when Indy
> finally matures and becomes "mainstream", people WILL notice the
> difference. Indy is already giving users more than some of the "deluxe"
> or "boxed" editions do, and doing it using GPL'ed software only.
>  

The Indy aproach is:

-Include what the user needs instead of what is flashy: allowing to
him to ask for help has ever been my obsession much more than KDE or
Gnome..

-Don't include software who would cause problems either technical or
 legal/philosophical.  As exalmples the fact install doesn't hhandle
 conflicts is the reason we have only one MTA and one news server
 despite my strong desire of providing an alternative to sendmail and
 INN.  As an example of software who is not included due to
 philosophical reasons there is Blender: the complete version is
 payware and I think Indy must remain neutral in payware wars ie not
 include crippled versions of it.

-Don't try to make the ultimate distribution but one who would help
 the user.  That implies not reiventing wheels just to make the
 prefect wheel if this implies users would have to fight linux bare
 handed for years until the ultimate distrib is ready

-Don't accept to be the puppet of another distrib.


> I'm looking forward to seeing the new release announcement on the
> Indy web site.
> 

About the people we need:

-Someone for propaganda, for making noise, for putting together the
 design notes I have been publishing from time to time.  This should
 be more democratic but the work who has been put in Indy is not all
 but mostly mine. :-) :-(

-A person for managing the people who will come when 0.2 will be
 announced.  Otherwise what will happen is: I will be too busy
 answering them for doing real work and they will see a project who is
 not moving.  This is will be bad for their morale so they will not
 help and this will be bad for my morale because Indy is an
 impossible mission for a single man.

> --
> 
> R.G. Mayhue rgmayhue@serenesoftware.com
> 
> ***********************************************************************
> 	The moral progression of a people can scarcely begin till they
> 	are independent.
> 						--James Martineau
> 
> 	Go get your Independence Linux now!
> 	http://www.independence.seul.org
> ***********************************************************************
> 
>