Linux-Development-Sys Digest #47, Volume #7 Thu, 12 Aug 99 21:14:10 EDT
Contents:
NFS performance with large files ("Walter F.J. Mueller")
Installing Oracle 8i on RED HAT LINUX 5.2 ([EMAIL PROTECTED])
Re: How do I use source RPMs (Robin Becker)
Re: Help ! Adaptec SCSI 2904CD ([EMAIL PROTECTED])
Re: Caldera kernel really 2.2.5?? (Alexander Viro)
Re: Best place in kernel for block encryption ? (Kaz Kylheku)
mmap corrupting system -- help! (Grant Edwards)
Re: Linux on HP 735 Workstation (Greg Fruth)
Proper Default Driver in Linux for HP IIIP (William Smith)
Re: threads (Sebastien Tanguy)
Re: RH 6.0 modprobe for lo (Johan Kullstam)
Device driver for Rockwell audio/modem card ("Xun Su")
Re: 2.2.x(?) problem with SO_LINGER... (Stephen Satchell)
Re: Best place in kernel for block encryption ? (Chris Gregory)
Device driver programming (root)
Re: Group ID maximum (Chris Gregory)
----------------------------------------------------------------------------
From: "Walter F.J. Mueller" <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.networking
Subject: NFS performance with large files
Date: Thu, 12 Aug 1999 18:09:18 +0200
Hi folks,
we are running here a couple of NFS servers and dual PII systems, still using
2.0.* kernels and NFS V2.
Some servers see at time high load, and some good fraction of the load is often
due to read/write and sometimes even deletion of large files. I consider a file
large if it has 100-500 Mbyte.
We have seen situations with pretty bad NFS response times (hangs of 10 sec
and more) and wonder, what the mechanisms behind are, and what possible
strategies to improve things could be.
I'm specifically suspicious about two things and appreciate input on that:
1. It seems, that deleting a large file locks down the whole file system
for a macroscopic time. For example, I rm'ed a *local* 500 Mb file in one
window and often, though not always, saw an `ls' on the same file system
block until the `rm' completed.
2. It seems, that `write' system are sometime blocked when a large amount
of cache buffers are flushed. For example, programs that write to a
*local* file with a large rate (e.g. 6 Mb/sec) often show a very
oscillarory CPU consumption, which is correlated with the flush
activities.
It seems, that in both cases a `global lock' gets in the way, and it's clear,
that such behaviour could result in bad NFS resonse times when some clients
handle large files while others expect good performance for small files.
* I wonder if somebody could comment on this ...
* Are there are any knobs available to improve this ?
* Are 2.2 kernels and NFS V3 going to address these issues ?
With best regards, Walter
--
Walter F.J. Mueller Mail: [EMAIL PROTECTED]
GSI, Abteilung KP3 Phone: +49-6159-71-2766
D-64291 Darmstadt FAX: +49-6159-71-2989
WWW: http://www-kp3.gsi.de/www/kp3/people/mueller.html
------------------------------
From: [EMAIL PROTECTED]
Subject: Installing Oracle 8i on RED HAT LINUX 5.2
Date: Thu, 12 Aug 1999 17:21:15 GMT
Hi
I am installing ORACLE 8i on RED HAT LINUX 5.2.
I had the following problem.could some one help me with it.
The "dba" group for oracle user was created and I continued with the
installation process. The installation process has to be run from the
start after the option create database account and filling the fields it
asks to locate the file for installation of product or in short call he
asks to select oracle product for installation. The first error came was
"failed to create unix.prd" file. After locating the file physically
copied thru the file manager utility the file unix.prd into the oracle
directory where it looks for this file during installation. After this
step was executed i had to again run the installer program, every time
the oracle installer program is run i had to fill the details again.
There was one more error at the product selection set screen this time
it could not find the file orainst.vrf in the hard disk. The error msg
was "orainst.vrf(0) : FILE_NOT_FOUND while verifying orainst. (no such
file or directory). Again i had to locate the file from CD and copy from
the CD to the hard disk in the specified directory where the
installation kit will be looking for it. After completing this process i
had to run again the oracle installer kit. This time it gave an error
msg while installing the products the error message was "oraview.vrf(0)
FILE_NOT_FOUND while verifying oraview. (no such file or directory)".
The file "oraview.vrf" was not there in both the CD's that you sent.
Therefore I could not further proceed as this file should be available
for proceeding with the installation process.
=======================================
Thanks
jim jagan
Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.
------------------------------
From: Robin Becker <[EMAIL PROTECTED]>
Subject: Re: How do I use source RPMs
Date: Thu, 12 Aug 1999 17:35:03 +0100
In article <fO4s3.2125$[EMAIL PROTECTED]>, Christopher Browne
<[EMAIL PROTECTED]> writes
>On 10 Aug 1999 16:19:29 -0400, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
>>But if you grab a source RPM and it doesn't build in your environment,
>>because it's different than the author's environment, then it's a pain
>>in the ass.
>
>It probably displays that either:
>a) The SRPM requires software that you haven't yet installed, which
>you need to know, or
>b) The specs weren't well-written, as it has some substantial
>environmental dependancy that was Dumb To Assume.
as an example modutils came with a bz2'd tar which RH 6.0 couldn't
handle until handomatically installed bzip2
--
Robin Becker
------------------------------
From: [EMAIL PROTECTED]
Subject: Re: Help ! Adaptec SCSI 2904CD
Date: Thu, 12 Aug 1999 17:49:31 GMT
And AHA-2940AU. I can't boot from that drive.
Thanks.
In article <[EMAIL PROTECTED]>,
Mathieu Lorentz <[EMAIL PROTECTED]> wrote:
> Does anyone know how and where can I find drivers for my Adaptec SCSI
> 2904CD
> I'm using a Redhat 6.0.
> Thanks
>
>
Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.
------------------------------
From: [EMAIL PROTECTED] (Alexander Viro)
Subject: Re: Caldera kernel really 2.2.5??
Date: 11 Aug 1999 21:20:30 -0400
In article <[EMAIL PROTECTED]>,
Michael Kahl <[EMAIL PROTECTED]> wrote:
[snip Caldera kernel is seriously different from the main tree]
>altogether and start clean with an officially released kernel? Will my
>system even still work if I do that (Caldera may have modified other
>components of the system to match their hacked-up kernel)?
IIRC they did - they used to include streams support <barf> into their tree
so probably they are using it somewhere.
>Do these problems exist with other distros as well? What's a poor kernel
>hacker to do?
Decide on which tree you are going to work? Debian, RedHat, Slackware
and SuSE userlands seem to be OK with the main tree. Let's see... Debian keeps
the patches in a separate file, so it should be the easiest to check... Aha.
Vanilla 2.2.10 + bigphysarea. So unless it has problems with the supplied
kernel it can't have any with the standard one. Checking the rest is left
as an exercise to reader. If Caldera ships seriously non-standard kernel
the basic decency would require to put the information on (in)compatibility
issues. Look around, maybe they have it somewhere...
--
"You're one of those condescending Unix computer users!"
"Here's a nickel, kid. Get yourself a better computer" - Dilbert.
------------------------------
From: [EMAIL PROTECTED] (Kaz Kylheku)
Subject: Re: Best place in kernel for block encryption ?
Date: Thu, 12 Aug 1999 16:53:35 GMT
On 12 Aug 1999 15:17:19 GMT, R�khar�ur Egilsson <[EMAIL PROTECTED]�encroute.fr>
wrote:
>Has there been any discussion here about where the most efficient
>place is in the kernel is to put an encryption layer.
>
>I would want it to work for all block devices. (i.e. all mountable
>filesystems, both current and future).
The proper place might be in the block device loopback driver. I believe
that this already exists.
------------------------------
From: grant@nowhere. (Grant Edwards)
Subject: mmap corrupting system -- help!
Date: Thu, 12 Aug 1999 20:18:39 GMT
I'm writing a simple driver for a PCI board that only supports
open, close, and mmap[*]. It has 3 minor numbers that mmap
different 3 differnt things:
1) A PCI memory range from a PCI board;
2) Another PCI memory range from that same board;
3) A page of memory allocated by the driver for use as a DMA
buffer.
The first two work fine. The third one sort-of works (the user
program sees the right page of memory), but it's corrupting
things (processes are croaking mysteriously, and the memory
management system complained about that page when shutting
down.)
I'm using a method described in "Linux Device Drivers" to mmap
a page of kernel memory by handling the page fault for the
desired page. I must be missing something, but I don't know
what. Rubini mentions incrementing the page usage count, but I
couldn't see that any drivers really do that (but then again,
they aren't mmaping unreserved memory).
Here's the mmap code:
======================================================================
unsigned scratchPage;
int init_module(void)
{
[...]
scratchPage = get_free_page(GFP_KERNEL);
[...]
}
/* handle page fault on scratch dma buffer page */
static unsigned long scratch_mmap_nopage(struct vm_area_struct * area,
unsigned long address,
int error_code)
{
int pos;
address &= PAGE_MASK;
pos = address - area->vm_start + area->vm_offset;
printk("mm nopage start=%08lx end=%08lx offs=%08lx\n",
area->vm_start, area->vm_end, area->vm_offset);
if (pos < 0 || pos >= PAGE_SIZE)
{
printk("invalid position = %d\n", pos);
return 0; /* seg fault user process */
}
printk("mm nopage returning %08x\n",scratchPage);
return scratchPage;
}
struct vm_operations_struct scratch_vm_ops =
{
NULL, /* open */
NULL, /* close */
NULL, /* unmap */
NULL, /* protect */
NULL, /* sync */
NULL, /* advise */
scratch_mmap_nopage, /* nopage */
NULL, /* wppage */
NULL, /* swapout */
NULL, /* swapin */
};
/* mmap for scratch dma buffer page */
int scratch_mmap(struct file * file, struct vm_area_struct * vma)
{
unsigned size = vma->vm_end - vma->vm_start;
printk("scratch_mmap() called with size=%d\n",size);
if (size != PAGE_SIZE)
return -EINVAL;
vma->vm_ops = &scratch_vm_ops;
return 0;
}
======================================================================
[*] I've been assigned to update the Linux device driver for a
board my company manufactures. The next generation will
include PCI bus-master DMA. Cool.
But first, I have to help troubleshoot the prototypes of
the new boards.
I've added some support for DMA to the driver, and I've
found a couple problems with the design of the DMA
controller. But, for various reasons, I can't easily do
much more troubleshooting with the standard device driver.
So, I want to mmap the device's two register spaces along
with a page for use as a DMA buffer. Then I can write a
test program in user space that runs the DMA controller
through it's paces.
--
Grant Edwards grante Yow! I think I'll do BOTH
at if I can get RESIDUALS!!
visi.com
------------------------------
From: [EMAIL PROTECTED] (Greg Fruth)
Crossposted-To: alt.os.linux,comp.os.linux.hardware
Subject: Re: Linux on HP 735 Workstation
Date: 12 Aug 1999 19:44:13 GMT
In article <[EMAIL PROTECTED]>, "Waldemar Born" <[EMAIL PROTECTED]> writes:
> Is there any version of Linux for HP Workstations??
Two answers: "Yes" and "Not Yet"
Answer #1: "Yes" (from the comp.sys.hp.hpux FAQ):
> Subject: 7.75+ Can I run Linux on HP PA-RISC hardware?
>
> Using MkLinux, yes.
>
> Take a look at http://www.osf.org/mall/os/pa-mklinux/
> for more information.
>
> Thanx to Rich Rauenzahn <[EMAIL PROTECTED]>
A newer link for MkLinux is: http://www.mklinux.org/
Answer #2: "Not Yet"
http://www.thepuffingroup.com/parisc/
MkLinux is based on the Mach microkernel, while the Puffin Group's effort
is based on the conventional monolithic kernel. MkLinux has been around
longer than the Puffin version (which is still in the very early stages of
development), but MkLinux/PA-RISC appears to have been orphaned.
(MkLinux/PowerPC is still alive, it seems.)
--
Gregory Fruth ([EMAIL PROTECTED])
The Aerospace Corporation
Los Angeles, CA
------------------------------
From: William Smith <[EMAIL PROTECTED]>
Subject: Proper Default Driver in Linux for HP IIIP
Date: Thu, 12 Aug 1999 13:56:31 -0600
Hello-
What's the proper default printer driver (in printcap) to use for the HP
IIIP? Tried the III drivers, but don't get the proper display for text.
Thanks,
Bill
------------------------------
From: Sebastien Tanguy <[EMAIL PROTECTED]>
Subject: Re: threads
Date: 12 Aug 1999 19:24:59 +0200
>>>>> "BB" == Bill Burris <[EMAIL PROTECTED]> writes:
BB> How are threads implemented on Linux? Are they part of the kernel
BB> or are they a user level library?
They are user level libraries around (MIT Threads...), but since at
least the 2.0 kernel we have kernel threads through the use of the
clone(2) system call (for example, the fork(2) call is roughly
equivalent to a clone(2) call with specifics parameters).
A POSIX (?) wrapper of this system call is known as LinuxThreads and
is included with the glibc. It offers usual functions like
pthread_create(...), etc...
BB> If they are implemented at the user level, what happens when a
BB> thread blocks on I/O? Do the other threads in the process still
BB> run, or is the complete process suspended?
Since they are kernel level threads, my guess is that a thread blocked
on I/O won't suspend other threads.
greetings,
seb.
--
Q: What's tiny and yellow and very, very, dangerous?
A: A canary with the super-user password.
------------------------------
Subject: Re: RH 6.0 modprobe for lo
From: Johan Kullstam <[EMAIL PROTECTED]>
Date: 12 Aug 1999 18:03:10 -0400
Robin Becker <[EMAIL PROTECTED]> writes:
> In article <[EMAIL PROTECTED]>, Johan Kullstam
> <[EMAIL PROTECTED]> writes
> I don't see that, but could it be related to searching for a nameserver.
> I don't have any eth so my net is local+default through ppp0
> >speaking of initrd, my disk access on kernel 2.2.x with x >= 7 on my
> >scsi system is also hosed (very slow, but no corrupt data. bizarre.)
> >unless i use an initrd. if have grovelled through many of the
> >initscripts but i cannot figure out this initrd dependence. until
> >then, i will just use initrd.
> I'm using standard 2.2.11 without any noticeable slowness. Are you
> fixing up the links in /boot as you go along?
yes. it happens only on the scsi box. and the effect is more
dramatic under smp.
here are some data points:
my simple machine
1 cpu, your vanilla ide hard drive
i can just drop in a kernel, run lilo and go
the big server machine
4 cpu, scsi hard drive, w/ built-in scsi driver, w/o initrd
kernel is 2.2.* where * >= 7
bonnie reports 40KB/sec read and write in single char mode.
using 1 cpu, scsi hard drive, w/ built-in scsi driver, w/o initrd
kernel is 2.2.* where * >= 7
i get 400KB/sec. an order of magnitude better.
4 cpu, scsi hard drive, w/ scsi driver as module, w/ initrd
any kernel
it's over 3000KB/sec.
4 cpu, scsi hard drive, w/ built-in scsi driver, w/o initrd
kernel is 2.2.* where * <= 6
over 3000KB/sec.
it's the damnedest thing and i have no idea what causes it.
euler(Bonnie)$ ./Bonnie
File './Bonnie.409', size: 104857600
Writing with putc()...done
Rewriting...done
Writing intelligently...done
Reading with getc()...done
Reading intelligently...done
Seeker 1...Seeker 2...Seeker 3...start 'em...done...done...done...
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
100 3136 97.4 4948 10.3 5371 26.1 3271 97.7 138819 100.3 8556.7 98.4
^^^^ ^^^^
these per char numbers are what i am refering to above.
the others are not affected as dramatically.
--
J o h a n K u l l s t a m
[[EMAIL PROTECTED]]
Don't Fear the Penguin!
------------------------------
From: "Xun Su" <[EMAIL PROTECTED]>
Subject: Device driver for Rockwell audio/modem card
Date: Thu, 12 Aug 1999 17:50:43 -0500
Is there anybody out there know if there exists a device driver for Rockwell
Riptide Audio/Communication PCI card. It is used in the HP Pavilion 6470Z pc
I am trying to configure into a linux box. Thanks for you all.
------------------------------
Subject: Re: 2.2.x(?) problem with SO_LINGER...
From: [EMAIL PROTECTED] (Stephen Satchell)
Date: Thu, 12 Aug 1999 18:14:47 GMT
> Linux kernel 2.0.x _did_ implement this feature, and so I expect it to
> be
> supported in 2.2.x. If it is no longer supported for some reason, as
> far as I can tell it is not documented.
>-James
>
Actually, I ran into this myself with 2.0.34, and haven't figure out how
to patch the old kernel to re-implement the SO_LINGER operation as
documented by Stevens. I have figured out how to get it back in the
newest kernels (2.2.5) so I'm getting new CD-ROMs and making the change.
Steven, later on in the book, really wails about doing this, saying that
"FIN is your friend" or something to that effect. Unfortunately, I've
discovered commercial implementations of TCP/IP that don't recognize a
FIN packet as an event worth interrupting the top-level application about
(like, for example, indicating that the socket should be read via SELECT
so that the poor app can know that there is a close attempt on the
socket) but a RST packet does the trick.
I understand why the Linux project would want to follow Steven's advice.
To whom do I send mail to suggest that the SO_LINGER thing be a kernel
configuration option?
------------------------------
From: [EMAIL PROTECTED] (Chris Gregory)
Subject: Re: Best place in kernel for block encryption ?
Reply-To: [EMAIL PROTECTED]
Date: Thu, 12 Aug 1999 23:59:56 GMT
On 12 Aug 1999 15:17:19 GMT,
R�khar�ur Egilsson <[EMAIL PROTECTED]�encroute.fr> wrote:
>Has there been any discussion here about where the most efficient
>place is in the kernel is to put an encryption layer.
>
>I would want it to work for all block devices. (i.e. all mountable
>filesystems, both current and future).
>
>I think one such place might be the cache buffers, any comments
>on that ? "linux/fs/buffer.c"
>
>
>--
> RIKHARDUR EGILSSON
> echo '[q]sa[ln0=aln80%Pln80/snlbx]16isb15CB32EF3AF9C0E5D7272C3AF4F2snlbxq'|dc
You could look at what people have done already. There's
linux-crypt-kernelpatches, which are for 2.0.11 kernel, or tcfs, which is
distributed with crypt patches for the kernel, I don't know what version.
Chris G.
------------------------------
From: root <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Subject: Device driver programming
Date: Wed, 11 Aug 1999 11:18:26 -0400
Hi,
I am new to Red hat Linux...Even though I have done device drivers
for NT and some embedded systems.. I am trying to write a dummy device
driver....Can somebody point to resources that can help me
Thanks for your help in advance
Jitendra Patel
OSS Corporation
------------------------------
From: [EMAIL PROTECTED] (Chris Gregory)
Subject: Re: Group ID maximum
Reply-To: [EMAIL PROTECTED]
Date: Fri, 13 Aug 1999 00:16:21 GMT
On Wed, 11 Aug 1999 01:47:27 +0200, Ralph Geissler <[EMAIL PROTECTED]> wrote:
>I have a user account on a Linux box (Intel) that has to be able to read
>a hundert user's directories without the users reading the directories
>of other users. I don't want my user account to be root.
>Every user is in it's own group and my user is in every single of these
>groups. Everything works fantastic. Except...
This is not really a good way to do this, because users generally own their
own directories. If one of them thinks to chmod go-rwx ~ then your plan will
fail. It's better just to do the read as root.
Chris G.
------------------------------
** 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
******************************
