Linux-Development-Sys Digest #644, Volume #6 Thu, 22 Apr 99 05:14:28 EDT
Contents:
Re: disable interrupt (Frank Sweetser)
Re: Maximum number of processes (Jay Thorne)
Re: Beowulf, Can we use it? (Christopher Browne)
Re: Beowulf, Can we use it? (Christopher Browne)
Re: EXT2 Sponsor File system (Christopher Browne)
Visual Code Comprehension Tools and ER Tools. ([EMAIL PROTECTED])
A Technical Question about the MBR Boot code. (John Smith)
Re: Can device driver use the filing system? (Villy Kruse)
Re: Monitor directory changes? (Alan Curry)
Re: Waiting for Red Hat 6.0 ("Folkert Meeuw")
Re: saving program state with core files (Matthias Warkus)
Re: I want Ctrl-Alt-Pause do a shutdown (system halt) (David Guyon Martin)
Re: Files larger than 2 GB on Intel/Linux (Steve D. Perkins)
Re: SCSI emulation for atapi devices (Bernd Strieder)
----------------------------------------------------------------------------
From: Frank Sweetser <[EMAIL PROTECTED]>
Subject: Re: disable interrupt
Date: 20 Apr 1999 12:40:29 -0400
"J.C. Chuang" <[EMAIL PROTECTED]> writes:
> Hi, all
>
> I often read some code that is used to disable interrupt by the
> following form:
>
> save_flags();
save the current interrupt state.
> cli();
clear interrupts.
> critical section....
>
> restore_flags();
restore the original interrupt state.
> What I don't understand is why no any code to enable interrupt
> again ??
picture this sequence of code.
some_func {
cli();
critical_stuff();
some_other_func {
cli();
critical_stuff();
sti(); /* restore interrupts */
}
more_critical_stuff(); /* oops - critical section w/interupts on! */
sti();
}
the save_flags/restore_flags pair ensures that when the function is done,
the interrupt state will be the same as it found it.
--
Frank Sweetser rasmusin at wpi.edu fsweetser at blee.net | PGP key available
paramount.ind.wpi.edu RedHat 5.2 kernel 2.2.5 i586 | at public servers
Odd that we think definitions are definitive. :-)
-- Larry Wall in <[EMAIL PROTECTED]>
------------------------------
From: Jay Thorne <[EMAIL PROTECTED]>
Subject: Re: Maximum number of processes
Date: Tue, 20 Apr 1999 17:21:02 GMT
James Stevenson wrote:
>
> Hi
>
> i belive (somenody correct me if am i wrong)
>
> you can have 32000 processes in the proc table (ish) should be a square
>
This is true for the software process table, but the hardware mm unit in
an intel is limited to 4096 processes. With a standard kernel and APM
enabled
you have 4090 processes available.
On alpha and PPC and sparc the limits are different (I forget what they
are)
I (HOPE!) believe merced is supposed to fix this.
--
Jay Thorne [EMAIL PROTECTED] KE Software
http://www.kesoftware.com
------------------------------
From: [EMAIL PROTECTED] (Christopher Browne)
Subject: Re: Beowulf, Can we use it?
Reply-To: [EMAIL PROTECTED]
Date: Thu, 22 Apr 1999 06:02:06 GMT
On Wed, 21 Apr 1999 14:44:30 -0700, Clint Byrum
<[EMAIL PROTECTED]> wrote:
>This is just a cut and dry, business application that is pretty
>disk-intensive. Everybody uses the same data, so that part is easy. I'm
>starting to think Beowulf isn't really what I need. Convince me that I'm
>wrong! It's got such a cool name. : P
If the issue is disk intensiveness, then methinks what you *really*
want is two things:
a) Get a pile of RAM. This allows reads to be cached a whole lot. If
what the users are going after involves reads and minimal writes, this
may be sufficient to give you really good results.
b) If the users update lots of data, then consider the "fast I/O" option:
i) Get the fastest SCSI controller you can get. Maybe two.
ii) Get several SCSI disks, and consider "RAIDing" the data across
several disks, and perhaps even across multiple SCSI buses.
A third thing; it is extremely preferable if there is able to be shared
memory access. If the application winds up dynamically allocating huge
amounts of memory for each user, that is a Bad Thing, as there won't be
any sharing of that. On the other hand, if they do memory-mapped access
of disk files, caches *will* be shared, which is a Very Good Thing.
In short, ample opportunities for tuning...
--
"And 1.1.81 is officially BugFree(tm), so if you receive any bug
reports on it, you know they are just evil lies." (By Linus Torvalds)
[EMAIL PROTECTED] <http://www.ntlug.org/~cbbrowne/lsf.html>
------------------------------
From: [EMAIL PROTECTED] (Christopher Browne)
Subject: Re: Beowulf, Can we use it?
Reply-To: [EMAIL PROTECTED]
Date: Thu, 22 Apr 1999 06:03:05 GMT
On Wed, 21 Apr 1999 17:16:06 -0400, G. Sumner Hayes <[EMAIL PROTECTED]> wrote:
>Clint Byrum wrote:
>> I'm trying to see if we can use Beowulf to provide bigger and/or more
>> robust linux systems to run our business apps on. Currently, we can
>> put(based on benchmarks) 200 users on a 4 Processor, intel SC450NX
>> Board w/ 1GB of RAM, a Mylex HW Raid Controller, running on the 2.2.2
>> kernel. What I want to know is, If we wanted to do 300 users, could
>> we just use beowulf to "cluster" in another machine of similar
>> caliber(or maybe half).
>
>Probably not. Beowulf applications need to be written to use PVM or
>MPI, so if you're using off-the-shelf business apps then it's not
>really a good choice. Moreover, it sounds like your application isn't
>really that well suited to a Beowulf-style approach in the first place.
Quite correct.
>> Would it be better just to get a Gigbit ethernet connect between them
>> and have one use nfs to access the other one(half the users would log
>> into the second machine)? Please advise! Thanks in advance!
>
>Some sort of load-balancing solution like this is probably simpler, but
>still not completely trivial if you want it to be transparent to the
>users. You could do round-robin DNS or something (so that logins to
>job.wherever.com get routed alternately to job1.wherever.com and
>job2.wherever.com based on whatever heuristic you want), but then you
>still have to deal with the regular distributed account problems
>(passwd changes, NFS or migrated home directories, NYS, etc). If you're
>already a distributed environment, fairly straightforward. If not,
>you have some reading to do.
>
>Coda will probably be an eventual solution to some of these issues, but
>it's not yet ready for deployment unless you're pretty familiar with
>it and ready to tinker and troubleshoot a fair amount.
>
>If you don't need a completely transparent solution, this becomes
>straightforward -- get a couple of boxes, figure out a way to route
>the incoming connections, and you're pretty much done. In theory.
These things describe the sorts of load balancing that would be done at
the network level, providing a different 'view' on parallelism than the
numerically-oriented stuff that Beowulf/PVM tend to provide.
The *third* option is to start looking at application-level parallelism.
For business applications, the database management and application logic
may be readily separable, so that you'd put the database on one "box,"
and then put the "application server" on another. This may help to
share the processing load across multiple machines, and rather parallels
the network-parallelism approach described above.
The fourth option that is good for heavy-duty "scaling" parallels the
PVM/Beowulf approach, and is to use some form of message-oriented
middleware (MOM) software to allow application functionality to be
serviced by many servers.
Tibco, Tecco, Tengah, OM3, Orbix OTM, and Tuxedo are the sorts of
products starting to be deployed on Linux that do this sort of thing.
A parallel to PVM/Beowulf is that the effective use of these products is
fairly likely to require redesigning the applications so as to decompose
the functionality into separable components.
The wide-sweeping generalization (that is pretty accurate) is that
there is no "silver bullet" of parallelism/clustering. Applications
must be suitably designed in order to take advantage of parallel
processing, and this requires rewriting "monolithic" applications.
--
"Oh, I've seen copies [of Linux Journal] around the terminal room at
The Labs." (By Dennis Ritchie)
[EMAIL PROTECTED] <http://www.hex.net/~cbbrowne/tpmonitor.html>
------------------------------
From: [EMAIL PROTECTED] (Christopher Browne)
Subject: Re: EXT2 Sponsor File system
Reply-To: [EMAIL PROTECTED]
Date: Thu, 22 Apr 1999 06:03:06 GMT
On Wed, 21 Apr 1999 17:02:12 +0200, Erik N. <[EMAIL PROTECTED]> wrote:
>This is how I think I will proceed from here:
>
>1. Make a new partition to hold the sp-fs.
> I believe this will involve modifying the source code for mke2fs (the
> tool for making a ext2 file system). And the default definitions of the
> ext2 directory structure.
I suggest that you add in step 0...
0. Establish a new identifier for the new filesystem, so that you can
fiddle with it however you like, and not get "YourFilesystem" partitions
mixed up with traditional ones.
I'd suggest calling it "ext9." ext2 is already taken by the common FS
that everyone uses. Stephen Tweedie is working on ext3, which should
provide some useful extensions (journalling/logging, and "big files," I
expect). ext9 probably minimalizes the confusion by making it obvious
that it's *your* custom filesystem.
--
"NT 5.0 is the last nail in the Unix coffin. Interestingly, Unix isn't
in the coffin... It's wondering what the heck is sealing itself into a
wooden box 6 feet underground..." -- Jason McMullan
[EMAIL PROTECTED] <http://www.ntlug.org/~cbbrowne/linuxkernel.html>
------------------------------
From: [EMAIL PROTECTED]
Subject: Visual Code Comprehension Tools and ER Tools.
Date: Thu,22 Apr 1999 20:45:59+2000
http://www.linuxcad.com
Yes this programmable visual modelling system can be used
for Visual Comprehension of code in any programming language and
for Entity Relationsip Modelling.
We also offer embedding of LinuxCAD in third party tools and
automation systems.
------------------------------
From: John Smith <[EMAIL PROTECTED]>
Crossposted-To: alt.linux
Subject: A Technical Question about the MBR Boot code.
Date: Tue, 20 Apr 1999 09:47:18 +0000
Hi...
I have a somewhat technical question regarding the boot code which I saw
at this URL:
http://metalab.unc.edu/navigator-bin/navigator.cgi?arch/i386/boot/bootsect.S
1) It says that the the BIOS start up loads the MBR code at address
0x7C00 -- this is the RAM memory address right?
2) It also says that the MBR code is moved from the address 0x7C00 to
the address 0x90000 -- I've been wondering for ages why it needs to move
the code. WHY? And why must it move it to the specific address 0x90000?
Why not some other address?
Well... those are all my questions. If possible, please include URLs or
documentation on how the MBR (Linux or any other OS is ok) sequence
works... It would be much appreciated. Thank you.
Kevin
------------------------------
From: [EMAIL PROTECTED] (Villy Kruse)
Subject: Re: Can device driver use the filing system?
Date: 22 Apr 1999 09:02:23 +0200
In article <[EMAIL PROTECTED]>,
Arun Sharma <[EMAIL PROTECTED]> wrote:
>
>In most cases, the problem can be solved by using a user level program
>which reads the files and does an ioctl to pass the data.
>
> -Arun
You could look at the loop driver which actualy does write to disk files
using the file system. There is, however, code in the kernel that takes
special considerations if the device drivers is the loop driver. If you
are not very careful when writing to a file from within a driver you might
very easily hang the system.
The loop driver is in /usr/src/linux/drivers/block/loop.c
Villy
------------------------------
Subject: Re: Monitor directory changes?
From: [EMAIL PROTECTED] (Alan Curry)
Date: Tue, 20 Apr 1999 18:02:37 GMT
In article <[EMAIL PROTECTED]>,
Sam Solon <[EMAIL PROTECTED]> wrote:
>Is there a (standard) way for a program to monitor a directory for
>changes and be notified when a change is made?
Unfortunately there isn't. The best you can do is periodically check its
mtime. Opendir it once and save the descriptor, and use fstat(dirfd(foo),&st)
and whenever st.st_mtime changes, rewinddir it and readdir the whole thing
again.
--
Alan Curry |Declaration of | _../\. ./\.._ ____. ____.
[EMAIL PROTECTED]|bigotries (should| [ | | ] / _> / _>
==============+save some time): | \__/ \__/ \___: \___:
Linux,vim,trn,GPL,zsh,qmail,^H | "Screw you guys, I'm going home" -- Cartman
------------------------------
From: "Folkert Meeuw" <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.networking
Subject: Re: Waiting for Red Hat 6.0
Date: Thu, 22 Apr 1999 07:49:53 +0200
Reply-To: "Folkert Meeuw" <[EMAIL PROTECTED]>
Nix <$}xinix{[email protected]> schrieb in im Newsbeitrag:
[EMAIL PROTECTED]
> Frank Sweetser <[EMAIL PROTECTED]> writes:
>
> > "folkert meeuw" <[EMAIL PROTECTED]> writes:
> >
> > > A month ago I updated to the Kernel Source 2.1.43.
> > ^^^^^^
But today I'm going back to 2.0.30 !
> >
> > ummm.... you *do* realize that not only is this an old beta version, but
Yes I realize, but supports more I understand now
> > this was right around the time that the filesystem rewrite actually
could
> > cause filesystem corruption?
Filesystem corruption was not the Problem !
The Problem was to routing with the Kernel, cause my PPP Connection don't
resets the default routing. More Problem was looking in the routing table
after
connect and see that is there only one PPP-Interface, but I need two for the
routing !!!
>
> That was 2.1.44 ;)
Please don't try to explain which release it is, I think that you show me I
need
more than only the Kernel source to "backe" a new .
MfG Folkert
------------------------------
From: [EMAIL PROTECTED] (Matthias Warkus)
Crossposted-To: comp.os.linux.misc,comp.os.linux.development.apps
Subject: Re: saving program state with core files
Date: Tue, 20 Apr 1999 18:25:44 +0200
Reply-To: [EMAIL PROTECTED]
It was the Tue, 20 Apr 1999 12:33:48 +0200...
..and dementen <[EMAIL PROTECTED]> wrote:
> Hi,
>
> I want to be able to save the state of a program (like a game for
> example) during its running and after to be able to load this state to
> continue the process at another time.
>
> I've tried with xxgdb 'name of the program' and then interrupting the
> execution but I don't know how to write the core file when the process
> is interrupted (it's not a beautiful way of doing what I want but it
> should be ok).
Core dumps are huge. I don't know whether all the state of a process
is in one, either. You don't want to create core dumps yourself.
Define your data structures well and write procedures or methods to
dump them into a file.
mawa
--
Look at the community. Look at how fast they turn the crank. They know
how to create a desktop environment out of the middle of nowhere in
less than two years - how could there be a company not scared by this?
-- mawa
------------------------------
From: David Guyon Martin <[EMAIL PROTECTED]>
Subject: Re: I want Ctrl-Alt-Pause do a shutdown (system halt)
Date: Thu, 22 Apr 1999 09:29:54 +0200
Thank you for your advice, but this url http://stekt.oulu.fi/~tosi/kbd/
doesn't work for me.
David
( remove .nospam.please from my email adress to answer )
Kalle Olavi Niemitalo a �crit:
> David Guyon Martin <[EMAIL PROTECTED]> writes:
>
> > I want Ctrl-Alt-Pause to do a shutdown (system halt).
>
> I have set Ctrl-Alt-Shift-Break to do the shutdown. Pressing it again
> after shutdown is complete will reboot the machine.
>
> This is done with the `loadkeys' program; see
> http://stekt.oulu.fi/~tosi/kbd/
>
> If you want to retain Ctrl-Alt-Del too (I didn't), you could bind
> Ctrl-Alt-Pause to KeyboardSignal (also known as Spawn_Console) and
> leave Ctrl-Alt-Del as Boot. Today's /sbin/init can catch both
> keypresses and handle them differently.
>
> Note that normal 102-key keyboards interpret Ctrl-Pause as Ctrl-Break,
> so you need to change the bindings of the Break key, not Pause.
>
> > Let me know it is a good idea and easy to do, I know C programming but
> > nothing about kernel development...
>
> No need to write C.
------------------------------
From: [EMAIL PROTECTED] (Steve D. Perkins)
Crossposted-To: comp.os.linux.misc
Subject: Re: Files larger than 2 GB on Intel/Linux
Date: Tue, 20 Apr 1999 18:27:54 GMT
Arghh... why does the time-stamp on your message have to be
the year 2018?!? Does this mean that this message will remain at the
end of the queue here for the next 19 years?
Steve
------------------------------
From: Bernd Strieder <[EMAIL PROTECTED]>
Subject: Re: SCSI emulation for atapi devices
Date: Thu, 22 Apr 1999 10:59:06 +0200
Hi
Marcin Romaszewicz wrote:
>
> Hi All,
>
> I have a question perhaps someone here can answer; I haven't found an
> answer in any docs.
>
> I've got an ATAPI CD-RW drive, and I am using cdrecord to burn CD's. In order
> for this to work, I must disable atapi cd-rom support and enable the scsi
> host adapter emulation, generic scsi support, and scsi cdrom support.
>
> Now, this is all well and good, and it works, I can successfully write CD's,
> but I can not mount CD's. To mount a CD, I either have to recompile the kernel
> or rmmod all the scsi stuff and insmod the atapi cdrom support.
>
> Is there a way, with scsi emulation and related modules installed, to mount
> an atapi cdrom as a scsi cdrom? When I try mounting /dev/cdrom (kernel says
> it detected a cdrom) or /dev/sda I get a message about an unsupported
> command getting sent to hdc (where the cdrom actually lives).
>
> Is there another way to mount the cd while the scsi support is present? I'm
> just looking to be able to burn a CD and mount one on the same kernel
> configuration.
>
> Any ideas?
> Thanks
Common problem:
Some distributions do something like > cd /dev; ln -s hdc cdrom; < where
hdc is the master on the second ide channel, the cd-rom in many cases.
They use /dev/cdrom to access the media to install al the things and the
system is configured to use /dev/cdrom.
But if you use scsi-emulation this moves as the previous posting says to
/dev/scd0 or /dev/scd1, so the link in /dev from cdrom to hdc is no more
valid, kill the link and use the true device-names by changing all
places where /dev/cdrom occurs, or link /dev/cdrom to the right device
avoiding confusion of your distributions configuration utilities.
If you have an ordinary cdrom and a cd-writer, I would recommend to set
the link /dev/cdrom -> device of cdrom (scd?), and use the true device
of the cd-writer (/dev/scd? or /dev/sg?? ), where appropriate
(configuration of xcdroast).
Perhaps it is wise to try to use your distributions configuration
utilities to remove the old atapi cd-rom and include the scsi-emulated
cdrom, perhaps everything is done then.
Bye,
Bernd Strieder
------------------------------
** 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
******************************
