minor corrections:
the whole "challenge" was a marketing ploy by oracle (not novell), aimed
at exercising some little used feature of both oracle and microsoft, but
a feature which oracle had spent a boatload of time optimizing solely
for this challenge. i don't recall the specifics of it, but it was
debunked by infoworld and other news agencies as well as microsoft.
i think oracle's "challenge" didn't restrict the hardware that could be
used either or even require it to be the same, so basically oracle was
allowed to run on a high-end sun and could use the results from that,
while microsoft had to run on wintel stuff. sorta like taking the state
champion high school football team and put them up against even the
lowly cleveland browns. the browns would eat them for breakfast in
their street clothes.
a very smart move by oracle - if microsoft refuses, well then obviously
they aren't as fast, and if they didn't, well, oracle is free to use
whatever numbers it can cook up in cahoots with sun and will blow msft
away no matter what they do.
however, this is all totally off-topic for the list and it's getting
really tiresome. please, take it to private mail, or set up a web page
where those of us with that kind of morbid fascination can follow along
with this challenge.
shag
=====
Judd Bourgeois | CNM Network +1 (805) 520-7170
Software Architect | 1900 Los Angeles Avenue, 2nd Floor
[EMAIL PROTECTED] | Simi Valley, CA 93065
Quidquid latine dictum sit, altum viditur.
----- Original Message -----
From: "Sam" <[EMAIL PROTECTED]>
Cc: <[EMAIL PROTECTED]>
Sent: Fri 25 Feb 2000 10:00
Subject: Re: Safecat challenge
> On Fri, 25 Feb 2000, Len Budney wrote:
>
> > Just for fun, why not put your money where your mouth is? I'll give
> > you $250 if you can write a program which 1) Exactly conforms to the
> > maildir protocol (including fsync()), and 2) runs 50 times faster
than
> > safecat, in an independent benchmark using actual emails, and 3)
runs
> > at least 20 times faster on a Linux machine with an ext2 filesystem
> > mounted async, like mine. Since this is such a lock for you, you can
> > dictate terms: what will you pay me after you fail?
>
> This reminds me of the "challenge" posed to Microsoft last year by, I
> think, Novell. They bet Microsoft $1,000,000 if they can prove that
the
> Microsoft SQL server is at least 1/100th as fast as their own database
> server. Of course, Microsoft did not take the bait, because their
> crapware runs only 1/20th as slow, and Novell would have gladly paid
> Microsoft a million dollars for the privilege of publicising that
> Microsoft admits that their own software runs 1/20th as slow as their
> competitors.
>
> So, basically, you want to be paid $250 for the privilege of writing
> inefficient code? Come up, 'fess up, do you work for Microsoft?
>
> Well, let's REALLY see how inefficient your code is, for which you
think
> you deserve $250 dollars. I happen to have a stripped maildir
delivery
> agent, which has been out there for six months, as part of
Courier-IMAP
> and SqWebMail. It's a little bit more that just a maildir deliverer,
it
> also calculates the current quota on the maildir, and the latest
version
> also supports sharable maildir folders, which involves additional
sanity
> checking to make sure that script kiddies aren't putting junk like
soft
> links in a public maildir. But, even with, I estimate, four or five
times
> more overhead than a completely basic maildir deliverer, looky what
> happens when the very message where you issue your <cough> "challenge"
> gets put through the ringer:
>
> [mrsam@ny safecat-1.1]$ cat t.c
> #include <stdio.h>
>
> int main(int argc, char **argv)
> {
> int i;
> int waitstat;
>
> for (i=0; i<1000; i++)
> if ( fork() == 0)
> {
> execv(argv[1], argv+1);
> exit(0);
> }
> else
> wait(&waitstat);
> return (0);
> }
> [mrsam@ny safecat-1.1]$ time ./t ./safecat tt/tmp tt/new </tmp/msg1
>
> [ snip ]
>
> 951500570.9689013621.ny.email-scan.com
> 951500570.0124413622.ny.email-scan.com
> 1.31user 2.03system 0:08.34elapsed 40%CPU (0avgtext+0avgdata
> 0maxresident)k
> 0inputs+0outputs (81074major+9017minor)pagefaults 0swaps
>
> [mrsam@ny safecat-1.1]$ time ./t
/usr/lib/courier-imap/libexec/deliverquota tt </tmp/msg1 ""
> 1.14user 2.31system 0:03.44elapsed 100%CPU (0avgtext+0avgdata
> 0maxresident)k
> 0inputs+0outputs (86075major+12017minor)pagefaults 0swaps
>
> So, even with me having four/five times of logical overhead (and I did
> manually add the final fsync to my code), I'm still 2.5 times ahead of
you
> on a small message. Most of the overhead here, certainly, is the
thousand
> forks and execs. Then, if we move this into the enterpise setting,
with
> PHBs sending 10 MB powerpoint presentations, I wish you the best of
luck
> writing that one, a byte at a time, ba-hahaha. And, that is very much
> real-life, real-world traffic that is common in many companies.
>
> Whether you like it, or not, your basic algorithm, the one that you're
so
> proud of, is much more than 50 times as slow as properly-implemented
> buffering algorithm.
>
> > > > I'll repeat, for the last time: safecat, on my system at least,
runs
> > > > within a factor of two of /bin/cat.
> > >
> > > And on other systems, it benchmarks at 1/50th the speed of
> > > efficiently-written code.
> >
> > 1. You benchmarked safecat? You saw somebody else's benchmarks? No.
So
> > shut up.
>
> I benchmarked its logic.
>
> I'm sorry to disappoint you with facts, but your code is very
inefficient.
>
> --
> Sam
>
>
>
