My test OpenBSD:

load from livecd bsdanywhere46-amd64

in different consoles:

dd if=/dev/wd0c of=/dev/null bs=1m
dd if=/dev/wd0c of=/dev/null bs=1m
dd if=/dev/wd0c of=/dev/null bs=1m
iostat
top

run:
dd if=/dev/urandom of=/dev/null
and disk read speed jump from 22Mb/s to 0.9Mb/s!

renice 20 for last dd - throughput not change!
renice -20 for first three dd  - throughput not change!
For check renice, run renice -20 for last dd - OpenBSD froze, even mouse.

It is not secure. One script or program may load CPU and 
database or another servers lost speed in disk operations.

In Linux (test on 2.6.35 libre) renice work correct...
Why renice not work in OpenBSD?

-- 
Dmitry Telegin

[ message continues ]

Then go to http://bsdanywhere.org/mailinglists and complain there.

Best
   Martin

[ message continues ]

My test very simple, please test your system. 
Is your original OpenBSD show identical results?


--
Dmitry Telegin

[ message continues ]

1) BSDanywhere is not OpenBSD

2) BSDanywhere is not existing anymore

3) It's live CD and it changes quite a lot things (eg. on which
controller is your CD and HDD ;-))

4) A lot of crypto inside OpenBSD http://www.openbsd.org/crypto.html
which is not true for Linux

5) where's your dmesg to see your HW?

6) Did you test it on real OpenBSD, real HW and latest release or snapshot?

Sounds like "professional" Linuxish style of "test" or "benchmark".




--
bIf youbre good at something, never do it for free.bB bThe Joker

[ message continues ]

http://bsdanywhere.org/faq
" What is the primary focus of BSDanywhere?
A mostly __unmodified__ OpenBSD kernel and userland...."


My test very simple, please test your system. 
Is your original OpenBSD show identical results?


--
Dmitry Telegin

[ message continues ]

I'm search stable and secure OS. 
I'm test: my work Mac OS X 10.6.3, FreeBSD 8.1 on livecd frenzy-1.3-ju-release-rus, my home Linux Debian - renice work more correct.
Why renice algorithm depends from hardware and why I need install OS on HDD? 
Is this renice behavior typical only for BSDanywhere and atypical for OpenBSD?

--
Dmitry Telegin

[ message continues ]

On Mon, 11 Oct 2010 11:27:45 +0400

I believe (if it's the right one) that it was meant to make it easier to

Even if renice did have that affect you need to be root to drop it to
-20 so why shouldn't root be able to use all resources, roots the only
user that can fill your disks (at default) completely too.

Security, stability and integrity are very different things often
working against each other. OpenBSD is definately the most secure by a
long way as that is it's primary goal and as it can go for long periods
easily without updates, likely the most stable. One OpenBSD specific
example of security and integrity contradicting each other is that the
security protections can actually make programs such as firefox more
likely to crash (arguably less stable) because of bugs in these
programs that are tolerated by other os. This leads to fixing these
bugs, making those programs more secure and stable. If you need the
latest flash content and the easiest install and maintenance, then

There's anonymos (old) And I have a couple of my own Openbsd livecd
here, one with a virus scanner on it but it's old now (3.8). There are
guides for building them that are easy to follow. But it won't be as
quick as you will probably like. Installing to hdd will be easier and a

[ message continues ]

I'm run dd if=/dev/urandom of=/dev/null _without_ renice and
disk read speed fell from 22Mb/s to 0.9Mb/s. Root with renice do not 


I will try to install.


-- 
Dmitry Telegin

[ message continues ]

use linux, you are clearly a moron, it will suit you better.

you have posted to this list about a dozen times in a 12 hour period, 
this is a sign you are an idiot. you complain about not being able to 
renice i/o, another sign. you are not even using openbsd on vmware, yet 
another sign you are an idiot. i'd say the case that you are an idiot is 
pretty well settled.

when you are using an OS you will rarely or never renice processes, it 
is a total waste of time. get a faster machine or let your machine sit 
and do its work. micromanaging a computer is a fool's errand.

[ message continues ]

your civility on this mailing list is decreasing by the day.
it was much better when you started.  perhaps now you feel

are you trying to tell me how should i use my computer?
that's a fool's erand indeed.

nice/renice has legitimate uses both on desktop and server.
it can really make the difference between a "business as usual"
and an almost unresponsive machine.

-f
-- 
unix should be used as an adjective. - at&t

[ message continues ]

after seeing all the crap this guy continued to post after i made my 
comment how can you possibly argue with my classification? i *totally* 
called it. plenty of other people became fed up with this guy after he 
continued to demonstrate his complete inability to grok attempts at 
educating him.

trolls abound on this list, you can count yourself amongst their 
numbers. why not harass any of the other people who, like me, were 
rightly critical of dmitry? you do it because you are a troll.

it's great to know you are professionally offended, i'm sure that must 


listen, it's not 1985 anymore, renice does not matter. if renice 
mattered someone would have fixed this long ago. i noticed this same 
behavior back in 2004 when i was screwing around with running cpu 
intensive simulations on an openbsd machine. after a bit of thought i 
realized that i was not using the machine correctly: i should not have 
services running on such a machine and expect it to be responsive when i 
slam the cpu.

[ message continues ]

i know, asking for a bit of civilness is trolling on misc@.


how about:  if you are feeling offended by the "sheer stupidity"
of a poster, and you are not interested in helping anymore
why not just send the name calling mails into /dev/null, hm?
or send a private mail.  it's not helping the archives, not
helping the guy and not helping anyone.  i call that trolling.


it is obvious the guy's english is on a barely usable level,
but so many people on this list take his attempts to communicate
as insult, impertinence, trolling..  here's a radical thought:
he cannot speak english.  or perhaps much younger than you
and is learning to communicate.  or a thousand other scenarios.
and yes, maybe he's just stupid.


normally, i would eat my own dogfood and ignore your
name calling mails to others.  i just thought that i chime
in as i see more and more mails where you say nothing useful
but slam people you have never seen or met.  not. cool.

-f
-- 
"fishing, stranger?"  "no, just drowning worms."

[ message continues ]

This is not how scheduling priorities work. You try to renice I/O bound
processes. The scheduler priority only matters when processes are CPU
bound.

I also don't get it why people always use /dev/*random as input
source for tests -- random numbers are not for free. 


[ message continues ]

Yes of course, but... all my  "dd" processes use CPU. 
After run "dd if=/dev/urandom of=/dev/null"
first three "dd if=/dev/wd0c of=/dev/null bs=1m" reduce their part of CPU and 
run "renice" not recover their part of CPU.

-- 
Dmitry Telegin

[ message continues ]

[ message continues ]

Write in your system 3 lines for numbers or GTFO

top
dd if=/dev/your_disk of=/dev/null bs=1m
dd if=/dev/urandom of=/dev/null

-- 
Dmitry Telegin

[ message continues ]

CPU consumed by the kernel is not accounted by the scheduler. All the
work done by urandom is system time.


[ message continues ]

And for the curious people who can't see the obvious: why is that?

-- 
Martin Pelikan

[ message continues ]

Because that's the way Unix and in particular BSD and its scheduler were
built. The kernel is assumed to be efficent and never doing lot of
computation. With the addition of crypto in the kernel this may no longer
be true but it does not affect normal operation. In other words nobody
ever bothered to fix this.


[ message continues ]

In FreeBSD and Mac OS X run "dd if=/dev/urandom of=/dev/null" 
not change read speed from disk...
Is in OpenBSD lacks developers?

-- 
Dmitry Telegin

[ message continues ]

Maybe.
But it doesn't lack moronic whiners on its mailing lists, that's for sure!

-david-

[ message continues ]

On Mon, 11 Oct 2010 15:34:22 +0400
 You need to think hard about your test and it's criteria. Often
 universities skew results to get funding, but you have no excuse.

It may be that the macosx and freebsd kernel gave more of a reason to
change this because their kernel is bloated and something hit them in
the face. I guess you haven't tried it on a proper install yet, either
but it would mean nothing anyway, unless you can crash the system as a
normal user.

All projects lack quality developers and OpenBSD needs quality!!
developers more than most projects.

Are there any problems caused by not having enough developers. As far
as I can tell No and I would think it would be far more likely that too
many developers of lesser quality would cause problems.

[ message continues ]

>>>>> "Dmitry-T" == Dmitry-T  <Dmitry-T@yandex.ru> writes:

Dmitry-T> Is in OpenBSD lacks developers?

That might as well be the last message you post here.

Any little help you would get, you've just offended them.

-- 
Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
<merlyn@stonehenge.com> <URL:http://www.stonehenge.com/merlyn/>
Smalltalk/Perl/Unix consulting, Technical writing, Comedy, etc. etc.
See http://methodsandmessages.posterous.com/ for Smalltalk discussion

[ message continues ]

reading his e-mails, I don't think he is trying to be offensive, I think 
his English is just poor

[ message continues ]

Thank you for your understanding.

-- 
Dmitry Telegin

[ message continues ]

I did not want to offend. Maybe the reason in my english.


-- 
Dmitry Telegin

[ message continues ]

It is strange... not well and not securely.
In Linux (test 2.6.35):
- run "dd if=/dev/urandom of=/dev/null" not change read speed from disk
- renice 20 for first three "dd if=/dev/wd0c of=/dev/null bs=1m" and
renice -5 for "dd if=/dev/urandom of=/dev/null" change throughput from 30Mb/s to 5Mb/s
- renice -20 for "dd if=/dev/urandom of=/dev/null" slows the system, but not freeze

-- 
Dmitry Telegin

[ message continues ]

Don't compare OpenBSD and Linux, because OpenBSD and Linux are made
differently. As claudio mentioned nice make sense *only* for CPU tasks,
not IO tasks. So if you don't like it keep using Linux, otherwise you
must live with how OpenBSD is doing things.


-- 
GP

[ message continues ]

I understand. 
How adminisrator in OpenBSD can give IO tasks more CPU?
This is very important for good control.


-- 
Dmitry Telegin

[ message continues ]

Read again and again Claudio's responses until you really understand. 
The key is there.

-- 
I will face my fear. I will permit it to pass over me and through me. 
And when it has gone past I will turn the inner eye to see its path. 
Where the fear has gone there will be nothing. Only I will remain.

[ message continues ]

thank you



-- 
Dmitry Telegin

[ message continues ]

On Sun, 10 Oct 2010 23:17:37 +0400

I have to admit that I was in the middle of a problem when I looked at
this, and got the wrong end of the stick (which process was at
-20). The email with the subject "insecure scheduler in OpenBSD 4.7"
was a much better description. Sorry for that Dmitry. However I do
think you would have had a better response if you had said something
like OpenBSDs service level gurantee is not as good with regard to disk
throughput than repeating openbsd isn't secure.

However I'm not sure that this is even the case as I have been unable
to repeat this problem on i386 without setting the normal user urandom
process to -20, at -5 for example it only uses 50% and the disks stay
the same. When all 4 are at 0 they have around 25% cpu each and
throughput is not affected.

Could this be an issue only for amd64? and possibly other architectures?
This test was done with vmplayer as it was quickest to run the test on.

Actually I've just noticed, for some reason the dd if=/dev/wd0c
of=/dev/null on it's own wants 100% cpu, so probably vmplayer is an
unfair test. Does the sheduler scale each process to a percentage of
what it wants or calls made? i.e 4 processes wanting 100% to 25% each
when they're all at the same nice level.

The interrupt is around 20% using wd0c and 5% using rwd0c.

[ message continues ]