Linux-Development-Sys Digest #716, Volume #6 Sat, 15 May 99 18:14:22 EDT
Contents:
Re: Destructive Erase? ("G. Sumner Hayes")
kernel past 2.2.5 failing to boot ([EMAIL PROTECTED])
Re: USB Support (John Thompson)
Re: How to make linux boot/shutdown rapidly ("Stefan Monnier "
<[EMAIL PROTECTED]>)
Re: Kernel-Thread in Linux? ("Stefan Monnier "
<[EMAIL PROTECTED]>)
Re: You can now use Winmodems in Linux!!!!!!! (Jonathan A. Buzzard)
Re: Ideas for ultra-high-performance web server (Phil Howard)
Re: Ideas for ultra-high-performance web server ("G. Sumner Hayes")
Re: Glibc rant (Nix)
Re: why top give false results? (Nix)
Re: NFS-Server setup (Nix)
Re: Linux disk defragmenter (M Sweger)
Re: 2.2.8 - Evil behavior (bryan)
Re: Reliable (!) nic for 2.2 kernel? (phil)
Re: 2.2.8 - Evil behavior ("G. Sumner Hayes")
Re: Strange behavior of SMP Linux 2.2.1 (Axel Farr)
----------------------------------------------------------------------------
From: "G. Sumner Hayes" <[EMAIL PROTECTED]>
Subject: Re: Destructive Erase?
Date: Sat, 15 May 1999 13:40:56 -0400
Tony Smith wrote:
>
> If you want this to be quick, use mmap() to map the file into virtual
> memory, and then you can use memset() or bcopy() to set the bytes to
> the values suggested by Bill.
If you do this, you may need to call msync() before fsync(); I don't
know what Linux's implementation is. If you're doing things securely,
it won't be any faster than the alternative -- generating good random
numbers (e.g. reading /dev/urandom) is the bottleneck. An extra
memory copy is just noise.
> bill davidsen wrote:
> > You open the file for read and write, seek to the end, get the
> > current position (aka the length), seek to the beginning, and then
> > write (length) bytes of zeros to the file. If you want security
> > against a pro physically taking apart your disk and getting bits off
> > under a microscope, write 0xFF, then zero, then 0xE5 to each byte,
> > three passes with a sync (or fsync) at the end of each write.
That won't protect very well against a determined attacker with
adequate resources. In DOD work, if a drive has ever held classified
data then it is considered classified for the rest of its working life.
It's amazing what you can find if you look at disk platters with an
electron microscope. But there are ways to do somewhat better (see
below).
> > Look for a program called 'purge' in the old comp.sources.unix
> > archives. I wrote it about 1991 and I believe I contributed it. I
> > got the 0xE5 from a security dude at DOE when I had clearance, don't
> > ask why that value, other than many format programs use it for the
> > initial low level format.
0xE5=10010101
It writes alternate bits well on drives that happen to use the proper
coding scheme (I forget which exactly; post-MFM/RLL but not the latest
drive tech). You're better off using cryptographically strong
random numbers. For a theoretical technical overview of why this is
true and what you're actually trying to do, see:
http://www.cs.auckland.ac.nz/~pgut001/secure_del.html
It's a bit dated, but still worthwile. I see that an update is planned
in the next couple of monts.
For a Linux implementation of the ideas in that paper, see:
http://users.erols.com/thomassr/zero/download/wipe/
Note the caveat for SCSI drives. Even if you disable write caching,
it's possible that there's block translation going on which could
leave some blocks undestroyed.
--Sumner
------------------------------
Date: Sat, 15 May 1999 12:25:48 -0400
From: [EMAIL PROTECTED]
Subject: kernel past 2.2.5 failing to boot
I've been running Alan Cox's patched 2.2.5 kernel for some weeks now and
decided to build kernel 2.2.9. The build was uneventful. However, the
kernel will not boot on my system. Right after identifying the IDE
devices, the system stops.
My main drive subsystem is SCSI (DPT 2044W) and only the primary IDE
controller is enabled. DPT BIOS is set to boot first. My system is SMP
and has all required updates. I also tried kernel 2.2.8 and had the
same problem so it must have been some change between 2.2.5 and 2.2.8
that caused this.
I can remember in the distant past the same kind of problem. However, I
cannot remember details nor can I find any information on this specific
problem.
Ideas and suggestions appreciated.
--
Shannon - [EMAIL PROTECTED] - www.widomaker.com/~shendrix/myresume.html
======================================================================
"All of us get lost in the darkness, dreamers turn to look at the
stars" -- Rush
------------------------------
From: John Thompson <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.misc
Subject: Re: USB Support
Date: Sat, 15 May 1999 08:14:12 -0600
Michael Proto wrote:
>
> Followup-To:
>
> On Sat, 15 May 1999 01:33:54 GMT, [EMAIL PROTECTED] muttered:
> >On 14 May 1999 17:16:55 GMT, Mark Hahn
> >>> <[EMAIL PROTECTED]> wrote:
> >>> "It can access the USB hub, and tell you what's on the hub. Actually
> >>> making use of the devices that are found may require drivers that
> >>> aren't There Yet..."
> >>
> >>that's a bit pessimistic.
> >
> >A bit, yes.
> >
> >>linux-usb people report that mice and kbds
> >>work fine enough to be depended on. there are sporadic reports of,
> >>for instance, working usb video cameras, and at least active development
> >>of storage devices.
> >
> >I see in the 2.8 kernel drivers for keyboards, that claims to be
> >workable, and a driver that *looks* like it's some sort of "multiport
> >serial thing." I didn't see mice, but may have misread something.
> >
> >At any rate, it *is* fair to say that what's there now is:
> >a) Somewhat beta-quality code for low level support; when the code's
> >so new it's rather early to call it "mature."
> >
> >b) *Some* device drivers.
> >
> >USB speakers are coming Real Soon Now; support for additional devices
> >will come on a device-by-device basis, and it is too early to predict
> >how quickly that will be.
> >
> >USB-ZIP drives are a neat option; who knows when?
> >
> >--
> >"You're one of those condescending Unix computer users!"
> >"Here's a nickel, kid. Get yourself a real computer" - Dilbert.
> >[EMAIL PROTECTED] <http://www.ntlug.org/~cbbrowne/lsf.html>
>
> IIRC, USB Zip drives *ARE* available. I've seen some that have a
> transparent-blue case that resembles a <shudder> iMAC.
Yes, they are available, but AFAIK they are not yet
supported by linux; hence the "who knows when" musing...
--
-John ([EMAIL PROTECTED])
------------------------------
From: "Stefan Monnier <[EMAIL PROTECTED]>"
<[EMAIL PROTECTED]>
Subject: Re: How to make linux boot/shutdown rapidly
Date: 15 May 1999 14:22:19 -0400
>>>>> "Konrad" == Konrad Mieredorff <[EMAIL PROTECTED]> writes:
> I'm wondering if it also saves the hardware state and/or reinitializes
> the hardware befor loading the image.
I'd expect that most of the boot process works just as before pretty
much until the hand is normally passed to init. Of course, I have no clue.
Stefan
------------------------------
From: "Stefan Monnier <[EMAIL PROTECTED]>"
<[EMAIL PROTECTED]>
Subject: Re: Kernel-Thread in Linux?
Date: 15 May 1999 14:25:56 -0400
>>>>> "Soohyung" == Soohyung Lee <[EMAIL PROTECTED]> writes:
> Does Linux(2.0.36) support Kernel-level Thread ?
You've looked really hard before asking this question haven't you ?
Stefan
------------------------------
From: [EMAIL PROTECTED] (Jonathan A. Buzzard)
Subject: Re: You can now use Winmodems in Linux!!!!!!!
Date: Fri, 14 May 1999 22:17:29 +0000
In article <7hckno$luk$[EMAIL PROTECTED]>,
[EMAIL PROTECTED] (bill davidsen) writes:
> In article <[EMAIL PROTECTED]>,
> Philip Boucherat <[EMAIL PROTECTED]> wrote:
>| What is a Winmodem anyway, and why would I want to use it?
>
> A winmodem is a device which uses a $200 CPU to substitute for a $4 UART
> in generating the clocked bit values for the modem output. This allows
> you to slow your system performance, and your modem performance, while
> reducing the initial hardware cost a little and boosting the
> manufacturer's profit at your expense.
>
Shame that this is generaly not the case then. Most winmodems differ
because they don't provide a serial interface with a device to
process AT commands. Neither of the above are at all necessary for
any modem.
What the windows drivers provide is a virtual serial port that it
uses to intercept AT commands and prod the chipset to do the right
thing. The only problem is that the chipset manufactures refuse to
release the programming information for their chipsets. If they did
Linux drivers would follow.
JAB.
--
Jonathan A. Buzzard Email: [EMAIL PROTECTED]
Northumberland, United Kingdom. Tel: +44(0)1661-832195
------------------------------
From: [EMAIL PROTECTED] (Phil Howard)
Subject: Re: Ideas for ultra-high-performance web server
Date: Sat, 15 May 1999 19:02:05 GMT
On 14 May 1999 23:15:12 -0700 Sam E. Trenholme ([EMAIL PROTECTED]) wrote:
| In light of the recent Mindcraft benchmarks, and ZDnet's similar
| benchmarks, the need for a web server solution that can handle the
| ultra-high-traffic typical of benchmarks is needed for the Linux platform.
| The way I am approaching this problem is that I am hacking thttpd to give
| it the kind of features I need (I added virtual hosts on a single IP), and
| then using that for NT-vs-Linux benchmarks.
|
| Anyway, thttpd uses the select() model to handle multiple connections at
| the same time. The problem with the select() model is that, as I
| understand it, it is very difficult to get this model to scale across
| multiple processers in a SMP setup. Therefore, I need some way of doing
| this in Linux:
The select() model, especially with sockets, is difficult to scale up
even on a single processor. You either have to mess around with locks
or you have multiple processors or kernel threads racing to see who can
accept() the next connection first.
I've been thinking about going into the kernel and see if I can change the
listen/accept semantics. My initial idea is to have the kernel only wake
up one process for each incoming SYN. Then I'd use the multiple listener
scheme that Apache uses, without the locking.
| * Putting four of the same 100-base-t NIC in a Linux box.
|
| * Binding one processer to each NIC in a four-way SMP system, and have one
| thttpd process for each NIC/processer combination.
|
| (Note that, alas, I do not currently have access to a four-way SMP box to
| do this kind of testing on)
Nor do most people who would be making a decision between NT and Linux.
Microsoft seems to think that everyone in the worlds must be using big
4GB quad processor machines (and designs their OS accordingly) just
because they can afford them.
| I am wondering how this can be done with Linux. Mind you, I know this is
| not practical in the "real world", but I am merely thinking about how to
| set up a system that looks good in the kinds of benchmarks PHM types read.
I think it is still better to "attack" unrealistic bendmarks, rather than
waste time and money trying to beat a benchmark that doesn't match most
real world scenarios. A good bench test of OS vs OS is to run on the
full scale of configurations each OS claims to support. Try every power
of 2 from 16meg on up.
| To make dynamic content look really good, I plan on hacking thttpd to
| directly, in its C code, make the needed dynamic content. The goal: Have
| dynamic content benchmark numbers that dust anything else out there.
How about in a shared library?
| The goal is not to set up a practical system as much as it is to set up a
| "Yes, Linux eats NT for breakfast in the web serving benchmarks" web
| server.
Run it on 4meg RAM vs NT on the same machine. See who can even get their
bacon fried with that!
Seriously, try 16meg.
--
Phil Howard KA9WGN
[EMAIL PROTECTED] [EMAIL PROTECTED]
------------------------------
From: "G. Sumner Hayes" <[EMAIL PROTECTED]>
Subject: Re: Ideas for ultra-high-performance web server
Date: Sat, 15 May 1999 15:28:35 -0400
Phil Howard wrote:
>
> On 14 May 1999 23:15:12 -0700 Sam E. Trenholme wrote:
>
> | In light of the recent Mindcraft benchmarks, and ZDnet's similar
> | benchmarks, the need for a web server solution that can handle the
> | ultra-high-traffic typical of benchmarks is needed for the Linux
> | platform.
Try zeus. That's what Linus and Alan recommend as a fast web server.
Maximum performance is not a goal of the Apache group. Good performance
and usefulness is.
> | Anyway, thttpd uses the select() model to handle multiple
> | connections at the same time. The problem with the select() model
> | is that, as I understand it, it is very difficult to get this model
> | to scale across multiple processers in a SMP setup. Therefore, I
> | need some way of doing this in Linux:
>
> The select() model, especially with sockets, is difficult to scale up
> even on a single processor. You either have to mess around with locks
> or you have multiple processors or kernel threads racing to see who
> can accept() the next connection first.
The development series of kernels (from 2.3.1) has wake-one accept.
I thing Andrea Arcalengi had some wake-one patches for 2.2.8.
--Sumner
------------------------------
From: Nix <$}xinix{[email protected]>
Subject: Re: Glibc rant
Date: 15 May 1999 17:11:46 +0100
Peter Mutsaers <[EMAIL PROTECTED]> writes:
> However this is not the case (in incredible stupid thing IMO; at least
> they should have bumped the major revison number again to 7 for
> glibc2.1)
No, they definitely should not have done so.
The public interface is *unchanged* (except that functions have been
added) between glibc-2.0 amd 2.1 except for the size of the FILE
structure and the hiding of some internal functions.
I don't want to recompile everything to go to glibc-2.1; rebuilding one
or two things is much better IMHO.
--
/* I hate C so much... */ --- jwz, in driver/xscreensaver.c
------------------------------
From: Nix <$}xinix{[email protected]>
Subject: Re: why top give false results?
Date: 15 May 1999 18:21:09 +0100
[EMAIL PROTECTED] (Jacek Pop�awski) writes:
> shows false results... is it true, that background processes not
> count to load avg? what about top?
If they did not, then the loadavg would be entirely useless :)
The loadavg is nothing but the avg size of the run queue over the last
(1 5 15) minutes. But if a process spends most of its time blocked (eg X
processes, daemons, and mpg123's reader thread) then it will not count
very much in the loadavg.
Until procps-1.9.0, top's CPU usage stuff was especially bad, and (AIUI)
the CPU usage of a process is only totted up at the end of its
timeslice, so processes that yield early can really distort the
figures. This is under active discussion on linux-kernel, so going on
past performance this is likely to improve soon :)
--
/* I hate C so much... */ --- jwz, in driver/xscreensaver.c
------------------------------
From: Nix <$}xinix{[email protected]>
Subject: Re: NFS-Server setup
Date: 15 May 1999 18:30:13 +0100
Zoran Cutura <[EMAIL PROTECTED]> writes:
> The message you get indicates that the portmap deamon is not running!
RH5.0's nfs daemons seemed to come with a broken portmap; nothing could
see it.
I rebuilt it from sources, and it worked fine.
--
/* I hate C so much... */ --- jwz, in driver/xscreensaver.c
------------------------------
From: [EMAIL PROTECTED] (M Sweger)
Crossposted-To: comp.os.linux.advocacy
Subject: Re: Linux disk defragmenter
Date: 15 May 1999 20:49:20 GMT
Reply-To: [EMAIL PROTECTED]
: > It gets about 25-50% of the way thru, then complains it is out of
: > memory. I have 128 megs of memory for Msofts defrag program.
: How big is the file system? I've seen the same behaviour on a
: 4Gb volume, and had to run defrag about 8 times to defragment the
: complete disk.
It's only a 2Gig partition on the hard drive - Drive E: that is set aside
for the UMSDOS linux filesystem. I only run defrag on this partition.
--
Mike,
[EMAIL PROTECTED]
------------------------------
From: bryan <[EMAIL PROTECTED]>
Subject: Re: 2.2.8 - Evil behavior
Date: Sat, 15 May 1999 20:11:06 GMT
G. Sumner Hayes <[EMAIL PROTECTED]> wrote:
: bryan wrote:
: >
: > Rodger Donaldson <[EMAIL PROTECTED]> wrote:
: > : You're lucky. None of the 2.2.x kernels I've tried work on one of
: > : my UMC 486 PCI motherboards - the mouse fails on all of them. 2.0.x
: > : works perfectly.
: >
: > its ps/2, also? weird. wonder what's different...
: ps/2 mice in general work fine with 2.2.x. Linus has a ps/2 mouse,
: so you can be sure that it'd get fixed quickly if it broke. Maybe
: you have a bit of an odd ps/2 mouse? The driver did change somewhat
: from 2.0.x -> 2.2.x.
: Anything special about your mouse or keyboard?
its a logitech wheel mouse (retail, not oem). its also a USB mouse
with a built in (real) ps2 adapter. it worked perfectly with 2.2.5
and .6 and others just not .7
its also going thru a linksys active switchbox. that has been some
trouble before - the cpu inside the switcher gets easily confused,
sometimes, but with .7, it gets confused a lot.
--
Bryan
------------------------------
From: phil <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.networking
Subject: Re: Reliable (!) nic for 2.2 kernel?
Date: Sat, 15 May 1999 20:40:39 GMT
bryan wrote:
> my tulip card is totally unreliable. I can bring it down with an ftp
> xfer (local lan) at 10 or 100, in a minute or less. network hangs and
> will NOT be reset by software.
>
> with a T1 download, it can hang the network in a few hours. this sucks ;-(
>
> even a '/etc/rc.d/init.d/network stop; /etc/rc.d/init.d/network start'
> won't fix the card. only a HARD reboot will reset it.
>
> has anyone done any load testing on the 2.2 kernel and found a
> RELIABLE nic card they could recommend? one that stays up under close
> to full load on a local 10/100 lan?
>
> (btw, I now have several dec tulip cards for sale. I refuse to use
> them in 2.2 kernel boxes I have here - sigh.)
>
> --
> Bryan
I have a Etherpower 10 PCI card that has worked fine under Linux 2.0.36 for
many months. As soon as I tried to use it with Linux 2.2.5 (Redhat 6.0) I had
problems with the board. The link light on the hub comes on for about 25
seconds then goes off until I reboot the system. The computer is a HP
Pavilion 6370Z with a 350MHz Pentium II. The Etherpower card works fine under
Windows 98 so I don't think the problem is the card. I tried defining
TULIP_NO_MEDIA_SWITCH in "tulip.c" because the documentation suggested
something similar but it doesn't seem to help.
Phil
------------------------------
From: "G. Sumner Hayes" <[EMAIL PROTECTED]>
Subject: Re: 2.2.8 - Evil behavior
Date: Sat, 15 May 1999 17:37:01 -0400
bryan wrote:
> its a logitech wheel mouse (retail, not oem). its also a USB mouse
> with a built in (real) ps2 adapter. it worked perfectly with 2.2.5
> and .6 and others just not .7
Hmm. There are some keyboard changes in the 2.2.7 patch, and the ps2
mouse driver is tightly coupled with the kbd drivers. In particular,
some of the changes to linux/drivers/char/pc_keyb.c remove error
checking. I'd try restoring those checks and seeing if it works.
If you're not comfortable doing that, try running 2.2.6 but
"#define KBD_REPORT_ERR" at the top of pc_keyb.c and see if you get any
messages saying "Keyboard buffer overflow" or "Keyboard error" on
console or in your log files.
Hope this helps.
--Sumner
------------------------------
From: Axel Farr <[EMAIL PROTECTED]>
Subject: Re: Strange behavior of SMP Linux 2.2.1
Date: Sat, 15 May 1999 23:19:23 +0200
David Fox wrote:
>
> Axel Farr <[EMAIL PROTECTED]> writes:
>
> > Some further notes on this problem:
> >
> > When the calculation processes are finished (or killed), then the system
> > behaves "normal" again.
>
> The first thing I would do is see how it performs under 2.2.9.
After a short test I get a the following behaviour under 2.2.9:
CPU State shows 150% user , ~50% system
(Extrapolating linearly, the problem should be solved at 2.2.17 :-) )
Nop, I got such a behaviour some times before, and it might be a first
stage to what shows up after some more time of operation. Processes like
"xeyes" appear to eat up 20 or 30 % of CPU time, even if no mouse is
moved. Something is going wrong there.
BTW: Why does my "xosview" Release 1.5 (from Suse 6.0) fail to run under
2.2.5 and 2.2.9? Under 2.2.1 it runs.
Greetings, Axel Farr
------------------------------
** FOR YOUR REFERENCE **
The service address, to which questions about the list itself and requests
to be added to or deleted from it should be directed, is:
Internet: [EMAIL PROTECTED]
You can send mail to the entire list (and comp.os.linux.development.system) via:
Internet: [EMAIL PROTECTED]
Linux may be obtained via one of these FTP sites:
ftp.funet.fi pub/Linux
tsx-11.mit.edu pub/linux
sunsite.unc.edu pub/Linux
End of Linux-Development-System Digest
******************************
