Linux-Development-Sys Digest #608, Volume #8 Wed, 4 Apr 01 10:13:07 EDT
Contents:
Re: did anyone have problems with Daylight Savings changeover? (H�kon Alstadheim)
Re: TCP/IP socket buffering (Florian =?iso-8859-1?Q?Gro=DFe=2DCoosmann?=)
Re: KERNEL PANIC on Mandrake 7.2 (Florian =?iso-8859-1?Q?Gro=DFe=2DCoosmann?=)
Re: CDR to Scsi ([EMAIL PROTECTED])
Re: mysql or postgresql? which is better? (Tim Roberts)
Re: TCP/IP socket buffering ("Joseph A. Knapka")
Embedded Linux Device (Jan Pietrusky)
Re: cpu scheduling problem ("�ڵ���")
Re: cpu scheduling problem ("�ڵ���")
Changing Hashes in ReiserFS ("Neeko Neekosan")
Re: did anyone have problems with Daylight Savings changeover?
([EMAIL PROTECTED])
Re: Writing a Device Driver ("Niels Sterrenburg")
Re: smp question (Rolf Magnus)
How can I use the SCI inside linux module for DSM?? (khaled Ragab)
----------------------------------------------------------------------------
Subject: Re: did anyone have problems with Daylight Savings changeover?
From: [EMAIL PROTECTED] (H�kon Alstadheim)
Date: 04 Apr 2001 00:07:23 -0400
Rick Ace <[EMAIL PROTECTED]> writes:
> "H�kon Alstadheim" wrote:
>
> > Set your hardware clock to UTC, and change the appropriate config
> > things. If you don't want to do that, live with it, or boot windows
> > once after every change (i.e. once in fall and once in spring) . It
> > will adjust your hardware clock.
>
> Hmmm. Windows might be good for something after all :oP
Tongue in cheek and all that, but just to keep the record straight:
windows is wrong on insisting that the RTC^H^H^H Hardware Clock be in
local time. Linux can actually work with both UTC and local time in
the hardware clock.
As we all know local time changes an hour in each direction every
year. Also computers have been known to move. When these things happen
we do *not* want the date a file was last changed to be affected.
System time should be UTC.
The windows way works, sort of, if you only run one OS. If linux did
the same as windows , you might end up with a *TWO* hour adjustment
after having booted windows and then Linux (depending on how the phase
of the moon affects windows ability to set the DST bit in CMOS). :/.
You *can* actually get linux to do the same thing as windows. Just put
something like "hwclock --systohc" in your halt script, and create a
check in your boot script to see whether daylight saving changed while
the machine was off. Add or subtract an hour accordingly with hwclock
--date=. This boot script should of course connect to the net and
download the latest laws on daylight saving, and parse out the start
and end dates, but that's a different story.
P.S: If you have N installations on a given machine, DST adjustment
will obviously be N hours, when you've booted each OS once. Try
dual-booting two windows installations and see.
--
H�kon Alstadheim, Montreal, Quebec, Canada
------------------------------
From: Florian =?iso-8859-1?Q?Gro=DFe=2DCoosmann?= <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.networking
Subject: Re: TCP/IP socket buffering
Date: Wed, 04 Apr 2001 07:58:25 +0200
Hi,
hebre wrote:
> Does anybody know how to disable receive buffering?
TCP >>must<< do buffering, this is the definition, see below.
> When client sends sequense of packets to server and server starts to
> receive them after some timeout (when all packets have been sent) it gets
> all packets in one recv() call, and buffer contains these in concatenated
> state. I have no information about this packets' format and can't to
> separate them, but I have to do it.
TCP collects data until the last data send has been acknowledged up to the
window size. It sends then the new data as one chunk to the reader. The
packets may be repackaged by routers etc. but the first bytes, e.g. 250,
will arrive as one block.
Moreover, the packets don't have the push flag set under normal circumstances.
This flag signals the receiver to "push" the data immediately to the
user process. The flag is normally set by the sender when it sends the last
packet of the chunk mentioned above.
> Help, please!
If you REALLY want to use TCP you have to acknowledge each of your own
packets (this leads to answers and you know where you are) or to implement
push flags or other worse things. You either have to live with a heavy
speed limitation or with your own TCP manipulations. Another chance which
relates to all of your TCP connections is to reduce your window size. This
can sometimes be done by the user process but it's silly since you can't
set your partner's window size.
Use UDP with packet numbering and you don't have any of the above problems.
One good book is R. Stevens' "TCP/IP Illustrated, Volume I" as far as I can
remember.
Cheers, Florian
------------------------------
From: Florian =?iso-8859-1?Q?Gro=DFe=2DCoosmann?= <[EMAIL PROTECTED]>
Subject: Re: KERNEL PANIC on Mandrake 7.2
Date: Wed, 04 Apr 2001 08:06:37 +0200
Hi,
Noe Nieto wrote:
> Hello,
> The instalation process seemend to be succes.... but when i restarted my
> machine i got that screen before a radical hang up:
> .
> .
> .
> Kernel Panic: atempted to kill iddle task!
> In swaper tasl- not sincing.
No panic, this is the usual error reporting of a buggy driver or other
kernel resource which isn't detected properly.
Cut down your hardware drivers until your machine runs. Then replace
the driver.
At least SuSE 7.1 has reported such an error in my environment. The
Adaptec driver has produced it. I built a new monolithic kernel and
that's it.
This error is reported very often in different FAQs.
Cheers, Florian
------------------------------
From: [EMAIL PROTECTED]
Subject: Re: CDR to Scsi
Date: 4 Apr 2001 06:10:44 GMT
In comp.os.linux.setup Stuart <[EMAIL PROTECTED]> wrote:
> I have scsi-ide as a module along with some other scsi and usb storage
> options.
> I cant make my cdr drive /dev/hdd be /dev/scd0
Add to the lilo.conf the option "hdd=ide-scsi", this will tell
the controller to leave hdd alone.
See also the CDWriting-HOWTO
Davide
------------------------------
From: Tim Roberts <[EMAIL PROTECTED]>
Subject: Re: mysql or postgresql? which is better?
Date: Tue, 03 Apr 2001 23:31:12 -0700
<[EMAIL PROTECTED]> wrote:
>[EMAIL PROTECTED] wrote:
>> It depends on what you want to do. If all you want is good read
>> performance, mysql is better. But mysql doesn't do transactions
>> which is a major problem if you ever do concurrent updates.
MySQL now supports transactions.
>Even that's up in the air. postgresql outperforms mysql by a handy
>margin on many workloads. ...
>As a general rule, postgreql is faster and more robust, while using
>more resources.
I have to say your performance notes are exactly contrary to my experience.
Having done web sites on both, I vastly prefer postgresql because of it
better feature set and robustness, but MySQL has outperformed Postgresql in
every benchmark I've ever done. The margin has been as much as 10:1 in
cases where I didn't use transactions.
>It's definately worth benchmarking both for critical applications.
Now, THIS is the kind of excellent advice that Usenet so rarely sees...
--
- Tim Roberts, [EMAIL PROTECTED]
Providenza & Boekelheide, Inc.
------------------------------
From: "Joseph A. Knapka" <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.networking
Subject: Re: TCP/IP socket buffering
Date: Wed, 04 Apr 2001 06:42:33 GMT
Grant Edwards wrote:
>
> In article <[EMAIL PROTECTED]>, Lew Pitcher wrote:
>
> >>Does anybody know how to disable receive buffering?
>
> [using TCP for datagrams]
>
> Bummer. I'm stuck supporting legacy products that use TCP as a
> datagram service. Programs that use TCP as a datagram service
> are just plain broken. But when they're already fielded, I
> guess you've got no choice but try to support them.
Reading from a TCP stream is no more problematic than reading
from a file sequentially. Any stream, in order to be understood
by the receiver, is going to have to be structured in some
way, so why the fuss if that structure happens to consist
of packets N bytes long? Just prepend a message length to each
packet, or install delimiters, or whatever, and
be prepared to handle recv() failures gracefully. I can't
see how this is any worse (more difficult/less reliable) than
trying to implement reliable datagrams on top of UDP;
in fact the latter seems quite a lot easier to get wrong
in some non-obvious way that takes many person-hours to
track down. Please enlighten me.
-- Joe
>
> There are two things you can do:
>
> First, find whoever designed a datagram protocol on top of TCP
> and bitch-slap him into next week. The program still won't
> work right, but you'll feel better.
>
> Second, there are socket options you can set to control the
> size of recieve and transmit buffers. See socket(7) and look
> for SO_RCVBUF and SO_SNDBUF. Changing them may or may not
> help, but it's something to try.
>
> >Please correct me if I am wrong, but it appears that you are using TCP
> >sockets. If so, then there's nothing you can do but ensure that each
> >packet contains some sort of delimiter that the receiver will
> >recognize as an 'end-of-packet' signal. You see, TCP is a byte-stream
> >protocol, and doesn't recognize data constructs of any size other than
> >8bit octets. It's up to the sending and receiving applications to
> >agree on some standard message format or structure in order to manage
> >data larger than a byte.
>
> Exactly.
>
> Unfortunately, there seem to be plenty people who do a "hello
> network" program with TCP and notice that on one particular
> network configuration, for some values of N, a single write()
> of N bytes results in a single read() of N bytes. They design
> an entire damn protocol around the faulty assumption that TCP
> always works that way. Then they deploy the program in the
> "real world" where there are routers and WANs and whatnot, and
> the thing doesn't work.
>
> Too bad nobody ever implimented the SOCK_SEQPACKET socket type
> for IP. It would be darned handy.
>
> --
> Grant Edwards grante Yow! Yow! Are you the
> at self-frying president?
> visi.com
-- Joe Knapka
"It was just a maddened crocodile hidden in a flower bed. It could
have happened to anyone." -- Pratchett
// Linux MM Documentation in progress:
// http://home.earthlink.net/~jknapka/linux-mm/vmoutline.html
* Evolution is an "unproven theory" in the same sense that gravity is. *
------------------------------
From: Jan Pietrusky <[EMAIL PROTECTED]>
Subject: Embedded Linux Device
Date: Wed, 04 Apr 2001 09:34:29 +0200
Hello,
I would like to design a embedded linux board with ARM7TDMI. I use
enough FLASH and SD-RAM for my design. Does anyone know, how is the
memory map of the linux? Which region of the memory contain the external
peripherie? How execute the boot process, should I use a boot eprom(
similar bios at PC)?
Thanks for all hints.
Jan
--
---------------------------------------------------------------------
| Dipl.-Ing.(FH) Jan Pietrusky | Tel: +49 (0) 3677 678331 |
| Institut fuer Mikroelektronik- und | FAX: +49 (0) 3677 678338 |
| Mechatronik-Systeme | |
| Langewiesener Strasse 22, 98693 Ilmenau | |
|---------------------------------------------------------------------|
| MAIL/WWW: [EMAIL PROTECTED], http://www.imms.de/ |
---------------------------------------------------------------------
------------------------------
From: "�ڵ���" <[EMAIL PROTECTED]>
Subject: Re: cpu scheduling problem
Date: Wed, 4 Apr 2001 16:54:15 +0900
"Simon J" <[EMAIL PROTECTED]> wrote in message
news:99pvt5$gl8$[EMAIL PROTECTED]...
> Dear Every one:
>
> I have question about that is given 5 jobs follows:
>
> Job Arrival Time Burst Time
> 1 0 3
> 2 2 6
> 3 4 4
> 4 6 5
> 5 8 2
>
> using the Round-Robin Scheduling with time quantum 1
>
>
> 0 job1 1 job1 2 job2 3 job1 4 ?? 5
> |______
> |_______|_______|________|________|_____________________
>
> but there has a problem at time 4, when time reach 4, we
know
> job1 will swap out and at the same time
> job 4 will arrive to ready queue but job 2 is already
wating
> in ready queue to execution,
> now the question is which job (job 4 or job 2) will be
> execute at time 4 ??
>
>
> Thank for your help!!
In LINUX scheduling algorithm, RR scheduling is considered to be a real-time
process (in fact, privileged process is more appropriate, i think)....
If you look into the source code, you can find the fact that when a new
process enters runqueue, it is located at the tail of the queue.
As David Schwartz said, two process can never be put into a run-queue at the
same time...
so, it depends on the time they arrive...
(although very short time difference... *^^*;;)
------------------------------
From: "�ڵ���" <[EMAIL PROTECTED]>
Subject: Re: cpu scheduling problem
Date: Wed, 4 Apr 2001 17:00:13 +0900
One more...
In SMP system...
two process can never be put into a runqueue at the same time either...
unless you modify the kernel to use several run-queues...
Linux run-queue is a single-level queue...
(virtually it can be thought to be multi-level queue...
For real-time processes, 99 level,
for non-real-time processes, 1 level
total 100 level...
)
"Simon J" <[EMAIL PROTECTED]> wrote in message
news:99pvt5$gl8$[EMAIL PROTECTED]...
> Dear Every one:
>
> I have question about that is given 5 jobs follows:
>
> Job Arrival Time Burst Time
> 1 0 3
> 2 2 6
> 3 4 4
> 4 6 5
> 5 8 2
>
> using the Round-Robin Scheduling with time quantum 1
>
>
> 0 job1 1 job1 2 job2 3 job1 4 ?? 5
> |______
> |_______|_______|________|________|_____________________
>
> but there has a problem at time 4, when time reach 4, we
know
> job1 will swap out and at the same time
> job 4 will arrive to ready queue but job 2 is already
wating
> in ready queue to execution,
> now the question is which job (job 4 or job 2) will be
> execute at time 4 ??
>
>
> Thank for your help!!
>
>
------------------------------
From: "Neeko Neekosan" <[EMAIL PROTECTED]>
Subject: Changing Hashes in ReiserFS
Date: Wed, 04 Apr 2001 03:24:10 -0500
Hi,
Is it possible to change the Hash function for my ReiserFS root
partition? It's using the bad-performing "tea" hash, which was the
deafault a few months ago. But now the default is the much better
r5 hash.
there is a mount option for hashes, hash=[r5,rusapov,tea,detect]...
Can I just change it to R5 and have it work? Am I risking damage, or will
it just completely fail outright?
thanks - nick.
------------------------------
From: [EMAIL PROTECTED]
Subject: Re: did anyone have problems with Daylight Savings changeover?
Date: Wed, 4 Apr 2001 10:35:30 +0200
H�kon Alstadheim <[EMAIL PROTECTED]> wrote:
> Rick Ace <[EMAIL PROTECTED]> writes:
>> "H�kon Alstadheim" wrote:
>>
>> > Set your hardware clock to UTC, and change the appropriate config
>> > things. If you don't want to do that, live with it, or boot windows
>> > once after every change (i.e. once in fall and once in spring) . It
>> > will adjust your hardware clock.
>>
>> Hmmm. Windows might be good for something after all :oP
> Tongue in cheek and all that, but just to keep the record straight:
> windows is wrong on insisting that the RTC^H^H^H Hardware Clock be in
> local time. Linux can actually work with both UTC and local time in
> the hardware clock.
FreeBSD also. Just
touch /etc/wall_cmos_clock
and you are compatible with windows.
--
Michel Talon
------------------------------
From: "Niels Sterrenburg" <[EMAIL PROTECTED]>
Subject: Re: Writing a Device Driver
Date: Tue, 3 Apr 2001 14:06:47 +0200
The Win Driver for Linux (not windriver) is available (30 days trial) at
www.jungo.com
"Josef Allen" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]...
> Mr Sterrenburg,
> Where can I get a linux device driver tool kit
>
> Josef D. Allen
>
> On Tue, 20 Mar 2001, Niels Sterrenburg wrote:
>
> > I hope you get the information... but I'm afraid that they are like most
> > companies and that you only get in contact with sales persons(and they
> > really don't know anything offcourse).
> >
> > Good luck
> >
> >
> >
> > "David Findlay" <[EMAIL PROTECTED]> wrote in message
> > news:7jzt6.36860$[EMAIL PROTECTED]...
> > > I have email conexant telling them that I would like to write a Linux
> > > driver for the conexant HCF(controllerless) chipset for modems. I told
> > them
> > > I would be writing it under the gnu public license for hopeful
inclusion
> > in
> > > the Linux kernel, and asked for a device driver kit to be emailed to
me.
> > > What do you reckon my chances of getting the information I need are?
> > Thanks,
> > >
> > > David
> > >
> >
> >
> >
> >
>
>
------------------------------
From: Rolf Magnus <[EMAIL PROTECTED]>
Subject: Re: smp question
Date: Wed, 4 Apr 2001 15:39:31 +0200
Rick Ace wrote:
> [EMAIL PROTECTED] wrote:
>>
>> Tim Roberts <[EMAIL PROTECTED]> wrote:
>> > ([EMAIL PROTECTED]) wrote:
>>
>> >>For smp's the value smp_num_cpus is used alot.
>> >>I wonder what happens to that value when one of the cpus crashes -
>> >>assuming linux doesn't crash because of this event.
>>
>> > Exactly how, in your experience, would a CPU "crash"?
>>
>> > Processes crash. Hard disks crash. CPUs do not crash.
>>
>> For example the fan on the CPU ceases to spin, and the CPU gets hot and
>> stops working. This is very common, much more frequent in my experience
>> than hard disk crash.
>
> I'd guess that the kernel is not prepared to handle failures of
> this nature. If that's correct, a hung system is a likely result,
> making the question about the value in smp_num_cpus moot.
When trying to write a driver, I sometimes had a crash on one cpu that
stopped it. The other one still was working though.
------------------------------
From: khaled Ragab <[EMAIL PROTECTED]>
Subject: How can I use the SCI inside linux module for DSM??
Date: Wed, 04 Apr 2001 16:07:39 +0200
Hi Everybody,
Can any one advise me solving the following problem.
I want to use the distributed shared memory DSM inside the linux kernel.
My platform is some PC's running Mosix kernel version 2.2.17
communicated by Dolphin PCI-SCI adapter cards D310. I wrote a module
which use the PCISCI device functions, but I can not open this device.
I received the following error when I call insmod for my module :
Unable to handle kernel NULL pointer dereference at virtual address
......
Have any one an idea how can I open a file inside the kernel space. I
need to use the shared memory segments in kernel space.
Many thanks for your in anticipation
Yours,
Khaled.
[EMAIL PROTECTED]
------------------------------
** 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 by posting to the
comp.os.linux.development.system newsgroup.
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
******************************
