From: Donn Miller <dmmil...@cvzoom.net>
Subject: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/26
Message-ID: <37751575.44F346EA@cvzoom.net>#1/1
X-Deja-AN: 494193760
Content-Transfer-Encoding: 7bit
X-Accept-Language: en
Content-Type: text/plain; charset=us-ascii
X-Complaints-To: news@ffx2nh3.news.uu.net
X-Trace: ffx2nh3.news.uu.net 930420141 26953 63.65.159.101 (26 Jun 1999 18:02:21 GMT)
Organization: UUNET
Mime-Version: 1.0
NNTP-Posting-Date: 26 Jun 1999 18:02:21 GMT
Newsgroups: comp.os.linux.advocacy
http://www.zdnet.com/pcweek/stories/news/0,4153,1015266,00.html
I'd like to thank Linux for being the sacrificial lamb, and
taking the spotlight away from FreeBSD. That's the thing about
running a hugely popular OS like Linux -- people start to attack
you from all sides. Maybe it's good FreeBSD is less popular?
Even though this is Linux and not FreeBSD, this whole thing about
"NT is better" is starting to worry me. It looks like NT, even
if it is better, is only better in the Networking only
categories. What about other intangibles like stability, uptime,
and configurability?
Sh*t, even if NT is better, where do you all suppose the
networking code came from? Maybe the title of the article should
be "NT beats Linux with Networking code they licensed from
Berkeley". God only knows what other pieces of code MS stole
from other OSes.
FreeBSD might fare better in such tests, but the bottom line is
MS is starting to compare their megabuck software against free
software, especially free UNICES. Tell me which OS has more bang
for the buck?
BTW, I find it funny that MS's IIs server supposedly "beats"
Apache in a lab test, but yet every web site running this
overrated piece of garbage has given me problems at one time or
another, but yet Apache-run web sites that I visited seemed more
reliable. IIs may be faster in a short block of time (as in a
lab test), but try measuring performance of the two over a long
block of time, and see which is more reliable. Again, there has
to be a reason so many web sites run Apache with Linux over
NT/IIs.
Another thing: the big joke is that NT is supposedly waaay
better, but compare the number of servers MS is using for FTP,
HTTP compared to what WC is running. If NT is so scalable, then
why does MS use >126 servers to run their "super scalable OS and
web server" on? In lab tests NT is better, but in the real
world, Microsoft loses.
So why is MS putting this crap in front of our faces? Does it
look I care if NT is better? NT can be 10,000,000 times better
than any UNIX as far as I'm concerned, but I'm not taking out a
loan so I can buy NT.
--
Donn
dmmil...@cvzoom.net
From: t...@y.jdyson.com (John S. Dyson)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/26
Message-ID: <7l36b6$30qr@enews5.newsguy.com>#1/1
X-Deja-AN: 494211245
References: <37751575.44F346EA@cvzoom.net>
Content-Type: text/plain; charset=us-ascii
Organization: http://extra.newsguy.com
Mime-Version: 1.0
Newsgroups: comp.os.linux.advocacy
In article <37751575.44F34...@cvzoom.net>,
Donn Miller <dmmil...@cvzoom.net> writes:
>
> So why is MS putting this crap in front of our faces? Does it
> look I care if NT is better? NT can be 10,000,000 times better
> than any UNIX as far as I'm concerned, but I'm not taking out a
> loan so I can buy NT.
>
Benchmarks seldom effectively measure real-world performance. Please
refer to my previous rants about lmbench being used as some kind
of expose on the vast improvement that Linux is over BSD.
IMO, benchmarks are best used for quality control purposes, and
for those who actually understand the limitations of the benchmarks.
Just looking at simple timing numbers (even if statistically well
controlled) don't often measure the actual performance under loading
conditions.
--
John | Never try to teach a pig to sing,
dy...@iquest.net | it makes one look stupid
jdy...@nc.com | and it irritates the pig.
From: Chris Costello <ch...@holly.dyndns.org>
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/26
Message-ID: <_d9d3.109$Gf.5660366@interramp>#1/1
X-Deja-AN: 494206129
References: <37751575.44F346EA@cvzoom.net>
User-Agent: tin/pre-1.4-980818 ("Laura") (UNIX) (FreeBSD/4.0-CURRENT (i386))
NNTP-Posting-Date: Sat, 26 Jun 1999 14:46:18 EDT
Newsgroups: comp.os.linux.advocacy
Donn Miller <dmmil...@cvzoom.net> wrote:
> I'd like to thank Linux for being the sacrificial lamb, and
> taking the spotlight away from FreeBSD. That's the thing about
> running a hugely popular OS like Linux -- people start to attack
> you from all sides. Maybe it's good FreeBSD is less popular?
This is an insightful article.
I seriously agree with you, here. FreeBSD aims to be the best
*server*, not the best desktop OS.
It's got what it takes to be a high end server for Alpha or
x86. ftp.cdrom.com is the prime example. More mature code such
as BSD will prove more stable and more reliable.
--
Chris Costello
Your password is pitifully obvious.
From: torva...@transmeta.com (Linus Torvalds)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/26
Message-ID: <7l3a35$kv6$1@palladium.transmeta.com>#1/1
X-Deja-AN: 494216471
References: <37751575.44F346EA@cvzoom.net> <7l36b6$30qr@enews5.newsguy.com>
Organization: Transmeta Corporation, Santa Clara, CA
Newsgroups: comp.os.linux.advocacy
In article <7l36b6$3...@enews5.newsguy.com>,
John S. Dyson <t...@y.jdyson.com> wrote:
>In article <37751575.44F34...@cvzoom.net>,
> Donn Miller <dmmil...@cvzoom.net> writes:
>>
>> So why is MS putting this crap in front of our faces? Does it
>> look I care if NT is better? NT can be 10,000,000 times better
>> than any UNIX as far as I'm concerned, but I'm not taking out a
>> loan so I can buy NT.
>>
>Benchmarks seldom effectively measure real-world performance. Please
>refer to my previous rants about lmbench being used as some kind
>of expose on the vast improvement that Linux is over BSD.
I still disagree, even when Linux lost.
Benchmarks are useful. They have to be distrusted, yes, but they _are_
useful.
For example, when we initially heard of the NT vs Linux numbers, pretty
much everybody just assumed that it was cooked. And yes, to some degree
it was cooked, but not by all that much. The re-run of the benchmark
clearly shows that NT _does_ actually kick ass on that benchmark, and
that's a valuable insight in itself.
Just ignoring the numbers (like some people in the Linux community do,
and like you always seem to do) is dangerous. They tell us something.
In particular, they told me that NT was better at this than I had
expected, and more importantly they showed a very definite weak area in
Linux.
No, the benchmark in question is not a very good one. It's not very
real world at all, but it _is_ interesting, the same way lmbench is
interesting, and the same way Spec95 is interesting to compiler people.
So John, instead of trying to explain away yet another set of numbers,
how about just 'fessing up to the fact that they quite often do mean
something. FreeBSD does equally badly in the comparison, and according
to some reports it does even worse. And this is the kind of benchmark
that FreeBSD should do well on, considering how people say that they try
to do a good server.
I'm not ashamed of bad numbers. Bad numbers are opportunities to make
them better.
But John, I _would_ be ashamed if I had to try to make up excuses for
them every time like you seem to do.
Linus
From: torva...@transmeta.com (Linus Torvalds)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/26
Message-ID: <7l3ajq$l7n$1@palladium.transmeta.com>#1/1
X-Deja-AN: 494218984
References: <37751575.44F346EA@cvzoom.net> <_d9d3.109$Gf.5660366@interramp>
Organization: Transmeta Corporation, Santa Clara, CA
Newsgroups: comp.os.linux.advocacy
In article <_d9d3.109$Gf.5660366@interramp>,
Chris Costello <ch...@holly.dyndns.org> wrote:
>Donn Miller <dmmil...@cvzoom.net> wrote:
>> I'd like to thank Linux for being the sacrificial lamb, and
>> taking the spotlight away from FreeBSD. That's the thing about
>> running a hugely popular OS like Linux -- people start to attack
>> you from all sides. Maybe it's good FreeBSD is less popular?
>
> This is an insightful article.
>
> I seriously agree with you, here. FreeBSD aims to be the best
>*server*, not the best desktop OS.
Oh, and that explaines why it does badly on a pure server test?
You do not make sense. This was _exactly_ the kind of test that a
server OS should do well on. Surprisingly, NT does well on it. Linux
doesn't. Neither does FreeBSD. Stop waffling about it, just accept it.
> It's got what it takes to be a high end server for Alpha or
>x86. ftp.cdrom.com is the prime example. More mature code such
>as BSD will prove more stable and more reliable.
It's fairly low-end as a server. "ftp.cdrom.com" is maxing out the
internet connection, but the benchmark that really stood out was the
file serving one. That's a case where you can't just dismiss out of
hand the fact that they used fast networking etc - fileserving is quite
often done on local networks, and the bandwidth is _much_ higher than
ftp.cdrom.com has ever seen. ftp.cdrom.com is a toy in comparison in
that way (it's impressive as a ftp-server with not all that much
hardware, don't get me wrong).
So I don't think you should be so complacent. What you're saying is
that FreeBSD is perfectly ok as long as you have a slow internet
connection, and that you don't care about file serving. No?
And please, don't take this as a rant against FreeBSD. Linux does badly
too. I acknowledge that, but I also want to point out that once we
found out that the benchmark was real (which took some time, admittedly)
at least we didn't stand around patting ourselves on our collective
backs. We got down and we're trying to fix the problem.
Linus
From: Chris Costello <ch...@holly.dyndns.org>
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/26
Message-ID: <stbd3.111$Gf.5725954@interramp>#1/1
X-Deja-AN: 494240756
References: <37751575.44F346EA@cvzoom.net> <_d9d3.109$Gf.5660366@interramp>
<7l3ajq$l7n$1@palladium.transmeta.com>
User-Agent: tin/pre-1.4-980818 ("Laura") (UNIX) (FreeBSD/4.0-CURRENT (i386))
NNTP-Posting-Date: Sat, 26 Jun 1999 17:19:20 EDT
Newsgroups: comp.os.linux.advocacy
Linus Torvalds <torva...@transmeta.com> wrote:
>> I seriously agree with you, here. FreeBSD aims to be the best
>>*server*, not the best desktop OS.
> Oh, and that explaines why it does badly on a pure server test?
I said it aims to. It's still in development as well as Linux
is. FreeBSD itself is roughly 2 years younger than Linux, isn't
it?
> You do not make sense. This was _exactly_ the kind of test that a
> server OS should do well on. Surprisingly, NT does well on it. Linux
> doesn't. Neither does FreeBSD. Stop waffling about it, just accept it.
I'm still interested in seeing the type of results FreeBSD
would get against NT or against Linux in the exact same
conditions.
I've seen more people that do accept that Linux didn't do as
good as NT on the specific test. In fact, they were glad that it
was done, because then they could analyze the results and see
what specifically was 'wrong' (and by wrong I mean 'can be
improved upon') and fix it.
>> It's got what it takes to be a high end server for Alpha or
>>x86. ftp.cdrom.com is the prime example. More mature code such
>>as BSD will prove more stable and more reliable.
> It's fairly low-end as a server. "ftp.cdrom.com" is maxing out the
> internet connection, but the benchmark that really stood out was the
> file serving one. That's a case where you can't just dismiss out of
> hand the fact that they used fast networking etc - fileserving is quite
> often done on local networks, and the bandwidth is _much_ higher than
> ftp.cdrom.com has ever seen. ftp.cdrom.com is a toy in comparison in
> that way (it's impressive as a ftp-server with not all that much
> hardware, don't get me wrong).
> So I don't think you should be so complacent. What you're saying is
> that FreeBSD is perfectly ok as long as you have a slow internet
> connection, and that you don't care about file serving. No?
No. I'm saying that the existing code from BSD, which is
rather mature, should prove more stable and reliable for file
serving than Linux or Windows NT. However, admittedly, I have
not looked at the Linux source code in a while -- does it
implement a sendfile(), which is a 'zero copy' system call that
sends a file (see http://www.freebsd.org/cgi/man.cgi?sendfile)
to a socket?
What I specifically meant by wcarchive being the prime example
is that FreeBSD can handle transfers (with large help, I imagine,
from sendfile(2)) of 1+ terabytes per day on a low-end
architecture.
> And please, don't take this as a rant against FreeBSD. Linux does badly
> too. I acknowledge that, but I also want to point out that once we
> found out that the benchmark was real (which took some time, admittedly)
> at least we didn't stand around patting ourselves on our collective
> backs. We got down and we're trying to fix the problem.
I'd also like to see specific tests of FreeBSD against NT in
those conditions, as I've said above and in different articles.
> Linus
--
Chris Costello
Diagnostics are the programs that run when nothing else will.
From: t...@halcyon.com (Tim Smith)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/27
Message-ID: <7l5ne9$iop$1@52-a-usw.rb1.blv.nwnexus.net>#1/1
X-Deja-AN: 494485598
References: <37751575.44F346EA@cvzoom.net> <7l36b6$30qr@enews5.newsguy.com>
<7l3a35$kv6$1@palladium.transmeta.com>
X-Trace: brokaw.wa.com 930504976 2792 206.63.251.52 (27 Jun 1999 17:36:16 GMT)
Organization: Institute of Lawsonomy
NNTP-Posting-Date: 27 Jun 1999 17:36:16 GMT
Newsgroups: comp.os.linux.advocacy
Linus Torvalds <torva...@transmeta.com> wrote:
>Just ignoring the numbers (like some people in the Linux community do,
>and like you always seem to do) is dangerous. They tell us something.
>In particular, they told me that NT was better at this than I had
>expected, and more importantly they showed a very definite weak area in
>Linux.
Have you read Helen Custer's "Inside Windows NT"? As far as I can tell the
suckiness of NT comes from two factors:
1. The parts outside the kernel that make is look/feel like Windows.
2. The fact that its not Unix, so it is unfamiliar to most kernel people.
The kernel itself appears to be very well designed, and implemented by
people with a lot of experience who do know what they are doing. Thus,
I don't see what people are surprised when it does some things well.
In fact, if NT weren't closed source, it would be a damned attractive
OS for a lot of things. E.g., it's really nice the way drivers in NT
can do most of the things applications can do (e.g., access files and
other drivers, create kernel threads)--combine that with a very layered
I/O system that lets you hook drivers in anywhere in there, and you can
do some pretty neat things.
--Tim Smith
From: torva...@transmeta.com (Linus Torvalds)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/27
Message-ID: <7l5pgs$n2o$1@palladium.transmeta.com>#1/1
X-Deja-AN: 494488071
References: <37751575.44F346EA@cvzoom.net> <7l36b6$30qr@enews5.newsguy.com>
<7l3a35$kv6$1@palladium.transmeta.com> <7l5ne9$iop$1@52-a-usw.rb1.blv.nwnexus.net>
Organization: Transmeta Corporation, Santa Clara, CA
Newsgroups: comp.os.linux.advocacy
In article <7l5ne9$io...@52-a-usw.rb1.blv.nwnexus.net>,
Tim Smith <t...@halcyon.com> wrote:
>
>Have you read Helen Custer's "Inside Windows NT"? As far as I can tell the
>suckiness of NT comes from two factors:
>
>1. The parts outside the kernel that make is look/feel like Windows.
>2. The fact that its not Unix, so it is unfamiliar to most kernel people.
>
>The kernel itself appears to be very well designed, and implemented by
>people with a lot of experience who do know what they are doing. Thus,
>I don't see what people are surprised when it does some things well.
Well, I have seen other benchmarks, and quite frankly NT does really
badly on them. NT has a lot of the same problems that a lot of
commercial UNIXes have: latency just sucks horribly on many common
operations.
For example, on lmbench the NT numbers are generally pretty pathetic.
And that's with a _native_ lmbench, not something that was working
through some POSIX emulation.
>In fact, if NT weren't closed source, it would be a damned attractive
>OS for a lot of things. E.g., it's really nice the way drivers in NT
>can do most of the things applications can do (e.g., access files and
>other drivers, create kernel threads)--combine that with a very layered
>I/O system that lets you hook drivers in anywhere in there, and you can
>do some pretty neat things.
People I consider trustworthy and in the know tell me that the NT
architecture just gets a lot of things wrong: getting the locking issues
in filesystems right is basically impossible etc.
It may have started out as a good design (but quite frankly I doubt it:
anybody who bases his design on VMS can't be all good), but it seems to
have grown uncontrolledly.
Linus
From: jer...@netcom.com (Jeremy Allison)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/28
Message-ID: <7l89pp$p7r@dfw-ixnews5.ix.netcom.com>#1/1
X-Deja-AN: 494832483
References: <37751575.44F346EA@cvzoom.net> <_d9d3.109$Gf.5660366@interramp>
<7l3ajq$l7n$1@palladium.transmeta.com> <stbd3.111$Gf.5725954@interramp>
Organization: Netcom
X-NETCOM-Date: Mon Jun 28 12:01:45 PM CDT 1999
Newsgroups: comp.os.linux.advocacy
NNTP-Posting-User: jeremy
Chris Costello <ch...@holly.dyndns.org> writes:
> I'm still interested in seeing the type of results FreeBSD
>would get against NT or against Linux in the exact same
>conditions.
Well according to Andrew (Tridgell) FreeBSD also did badly on
the file serving test using the same version of Samba (I was
in Paris when the FreeBSD guys did their tests at PC Week so
wasn't able to help or observe). Now before you point the finger
at Samba, note that the same Samba on Solaris beats NT (admittedly
with a RAM disk, but this benchmark essentially runs completely
in RAM anyway, due to the oplocks reducing the real working set).
It's the TCP stack. Am I correct in saying FreeBSD has a singly
locked stack on an MP machine, as does Linux ?
Regards
Jeremy Allison,
Samba Team.
From: t...@y.jdyson.com (John S. Dyson)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/29
Message-ID: <7l92lk$2a1h@enews3.newsguy.com>#1/1
X-Deja-AN: 494986331
References: <37751575.44F346EA@cvzoom.net> <_d9d3.109$Gf.5660366@interramp>
<7l3ajq$l7n$1@palladium.transmeta.com> <stbd3.111$Gf.5725954@interramp>
<7l89pp$p7r@dfw-ixnews5.ix.netcom.com>
Content-Type: text/plain; charset=us-ascii
Organization: http://extra.newsguy.com
Mime-Version: 1.0
Newsgroups: comp.os.linux.advocacy
In article <7l89pp$...@dfw-ixnews5.ix.netcom.com>,
jer...@netcom.com (Jeremy Allison) writes:
> Chris Costello <ch...@holly.dyndns.org> writes:
>
>> I'm still interested in seeing the type of results FreeBSD
>>would get against NT or against Linux in the exact same
>>conditions.
>
> Well according to Andrew (Tridgell) FreeBSD also did badly on
> the file serving test using the same version of Samba (I was
> in Paris when the FreeBSD guys did their tests at PC Week so
> wasn't able to help or observe). Now before you point the finger
> at Samba, note that the same Samba on Solaris beats NT (admittedly
> with a RAM disk, but this benchmark essentially runs completely
> in RAM anyway, due to the oplocks reducing the real working set).
>
> It's the TCP stack. Am I correct in saying FreeBSD has a singly
> locked stack on an MP machine, as does Linux ?
>
Some of the guys & Kirk are working on the locking issues. I suspect
that they are making progress (I am clued in, but am only passively
participating -- watching the private and public email traffic go by.)
(They are being kind to me keeping me up to date -- probably due to the
fact that I will be participating more soon again, but for now, the new
'blood' working on the code bodes very well.)
Most of the problem is due to the single lock, and likely will fix
two immediately apparent scalability issues. (It isn't just the
single lock of the TCP stack at issue.)
They have alot of work ahead of them in order to perform super well in
the Mindcraft benchmark, but I think that the individuals involved
are capable...
--
John | Never try to teach a pig to sing,
dy...@iquest.net | it makes one look stupid
jdy...@nc.com | and it irritates the pig.
From: jer...@netcom.com (Jeremy Allison)
Subject: Re: [OT] Why it's a good thing FreeBSD isn't as popular as Linux
Date: 1999/06/29
Message-ID: <7lb0ou$j67@dfw-ixnews11.ix.netcom.com>#1/1
X-Deja-AN: 495271691
References: <37751575.44F346EA@cvzoom.net> <_d9d3.109$Gf.5660366@interramp>
<7l3ajq$l7n$1@palladium.transmeta.com> <stbd3.111$Gf.5725954@interramp>
<7l89pp$p7r@dfw-ixnews5.ix.netcom.com> <7l92lk$2a1h@enews3.newsguy.com>
Organization: Netcom
X-NETCOM-Date: Tue Jun 29 12:46:06 PM CDT 1999
Newsgroups: comp.os.linux.advocacy
NNTP-Posting-User: jeremy
t...@y.jdyson.com (John S. Dyson) writes:
>In article <7l89pp$...@dfw-ixnews5.ix.netcom.com>,
>Some of the guys & Kirk are working on the locking issues. I suspect
>that they are making progress (I am clued in, but am only passively
>participating -- watching the private and public email traffic go by.)
>(They are being kind to me keeping me up to date -- probably due to the
>fact that I will be participating more soon again, but for now, the new
>'blood' working on the code bodes very well.)
>Most of the problem is due to the single lock, and likely will fix
>two immediately apparent scalability issues. (It isn't just the
>single lock of the TCP stack at issue.)
>They have alot of work ahead of them in order to perform super well in
>the Mindcraft benchmark, but I think that the individuals involved
>are capable...
This is good news indeed ! I don't care whether it's FreeBSD
or Linux underneath, so long as it's running Samba :-) :-) :-).
Cheers,
Jeremy Allison,
Samba Team.