Linux-Development-Sys Digest #601, Volume #6 Thu, 8 Apr 99 20:14:00 EDT
Contents:
Re: persistent heap design advice please (Phil Howard)
Re: Arrgghh! How MUCH does it cost to set up Apache? ("Walter B. Burke")
Re: Stream Processing (provided by SVR4) for Linux? (Andi Kleen)
Re: Stream Processing (provided by SVR4) for Linux? (David Grothe)
Re: Switchbox + mouse = braindead mouse? (Phil Howard)
Re: gdb + linuxthreads + kernel 2.2.x = fixed :) (Kaz Kylheku)
Re: 4 Gb memory? ECC? (Phil Howard)
Boot feature extension? ("D. Stimits")
Re: Interrupts are ok, but ports in serial.c? (Vipin Malik)
Ranish Partition manager ("D. Stimits")
Where can I get the e-mail for person responsible for ramdisk.c in kernel ? (Vipin
Malik)
Re: Where can I get the e-mail for person responsible for ramdisk.c in kernel ?
(Daniel Robert Franklin)
Re: meminfo and ps (Phil Howard)
----------------------------------------------------------------------------
From: [EMAIL PROTECTED] (Phil Howard)
Subject: Re: persistent heap design advice please
Date: Thu, 08 Apr 1999 22:24:10 GMT
On Thu, 8 Apr 1999 09:47:53 -0500 Keith Morgan ([EMAIL PROTECTED]) wrote:
| I am interested in creating a persistent heap library and would appreciate
| any hints or
| suggestions on how to proceed. The 'persistent heap' would be a region of
| virtual memory
| backed by a file and could be expanded or contracted.
|
| In order to build my 'persistent heap' it seems like I need a fundamental
| facility that isn't
| provided by Linux. Please correct me if I'm wrong! It would be something
| like a blend of
| the mmap() and kernel virtual memory facilities. The facility (call it
| phmap) would:
IWSTM, that everything you need is provided by Linux. I would bring into
question exactly how you want to contract the space. Would that contraction
only happen between instances of use?
By persistence between processes, if the first process did several calls
to access space, and the subsequent process also does several calls, how
do you intend to correctly associate the pieces of data from one to the
other? Will the pieces be associated by order of call, or will there be
some sort of identifier system?
Why not let each piece be a separate file? Or is the intent be sure it is
all self-contained ... that would imply some sort of directory inside the
file, kinda like a PKZIP file has (although with different needs).
| admit that I don't
| have a coherent picture of memory management yet. Thanks for any insight.
My first thought is that the model of memory that processes see via calls
like mmap() should be adequate, and probably essential for portability.
Speaking of portability, there would be the issue of how to handle such
data being transferred between architectures. One likely way to handle
it is to put an architecture code in the file and reject it if it does not
match. Something more abstract would be needed for portable data, but
that's why we have files. The definition for persistent heaps probably
should be that the facility will be available on all architectures, but
that the data cannot be transported between incompatible architectures.
--
Phil Howard KA9WGN
[EMAIL PROTECTED] [EMAIL PROTECTED]
------------------------------
From: "Walter B. Burke" <[EMAIL PROTECTED]>
Crossposted-To: linux.admin.isp,alt.comp.linux.isp
Subject: Re: Arrgghh! How MUCH does it cost to set up Apache?
Date: Thu, 8 Apr 1999 13:51:25 -0500
Why would you need any more than one IP address?
Clay Reiche wrote in message ...
I've set up an Apache web server in Florida using a 384kbs ADSL. The
phone company charged $99 for the installation AND the modem! Not bad! For
ADSL they charge based on the amount of band width you require. 384kbps
costs me $55 a month from the phone company. Then there's the ISP charge...
I shopped for a week to find the best rate... $199 a month and they gave me
a block of 64 static IP addresses!(Don't know what I'm gonna do with them
all, but I'm sure I'll find something...) They charged $49 for installation.
I found one company that was cheaper, $149 a month, but they only gave me
one IP address...(they would charge $5 per additional IP) and the
installation was $149! So, my monthly expense is $250 and the installation
expense was $104. I know that you don't live in Florida, but maybe this can
give you an idea...?
Clay
PS: I didn't even consider T1, prices are outrageous.(and I'm running
this out of my apartment.)
Wayne Chunn wrote in message <[EMAIL PROTECTED]>...
[EMAIL PROTECTED] wrote:
I've been poking around for a bit on the relevant newsgroups,
and
trying search engines, but can't seem to get any information at
all
about how roughly much it would run to start a small web hosting
service,
exclusive of the hardware. I mean some idea, *any* idea, of
current
rates in the Rochester, NY area or thereabouts, for T1, ASDL
(yeah, I
know -- but it might be okay if there were multiple lines,
perhaps
cheaper than the same capacity from a single T1 if I understand
the
racket the telephone companies are running on T1/T3 etc.), or
whatever.
The telephone company sites simply refuse to provide this
information,
and I've been through this sort of wild-goose chase enough times
to know
that if the first few conceptually related sites will not
provide
some piece of information, then they're *all* going to be hiding
that information, or will be wanting to play head games on you
before
finally grudgingly admitting to even some vague price or
another.
Fuck that crap. I'll read the FAQ, if I can ever find one, or
a book.
ANYTHING is better than a predatory salesman trying to probe my
mental
armour for a chink or flaw he can exploit to screw me for all he
can get.
Speaking of which, a FAQ pointer or book recommendation for
setting
up Apache under Linux, to run a small server network with
perhaps
four or five 200MHz Pentium class junkers with maybe only 64M
each,
would be very useful. :)
-----------== Posted via Deja News, The Discussion Network
==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start
Your Own
I can't help you much with your primary request (pricing) but for
your request in
the last paragraph I would suggest the Special Edition of Using
Linux Part 7, Chapters
35-37. Also, if you have access to several 200 MHz Pentiums you may
be interested
in experimenting with Parallel Processing. Search the web for info
on current PP Linux
projects if that interests you, there are quite a few.
Later,
Wayne
------------------------------
From: Andi Kleen <[EMAIL PROTECTED]>
Subject: Re: Stream Processing (provided by SVR4) for Linux?
Date: 08 Apr 1999 23:09:52 +0200
David Grothe <[EMAIL PROTECTED]> writes:
>
> I am supposed to believe that there are two parallel implementations of
> TCP/IP in Solaris? If so, what is the driver name of the "socket based
> TCP/IP"?
It seems though that Solaris uses some fast path through system STREAMS modules
if applicable (=all modules on the stack support it), and it is not clear
if you would recognize that as STREAMS, as you know it.
BTW, how is the LiS port to 2.2 coming along?
-Andi
--
This is like TV. I don't like TV.
------------------------------
From: David Grothe <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Subject: Re: Stream Processing (provided by SVR4) for Linux?
Date: Thu, 08 Apr 1999 22:28:52 GMT
Andi Kleen wrote:
> BTW, how is the LiS port to 2.2 coming along?
>
> -Andi
I have a candidate version of LiS here to be called version 2.0 (or maybe 2.1). It
works with 2.0 kernels and 2.2 kernels. An LiS aficionado has added very nice
loadable module support for STREAMS drivers -- load on demand stuff. Also a DLPI
interface to the standard Linux MAC drivers (Ethernet, etc). I am going to augment
that latter feature with an interface to the Linux Token Ring driver.
I will be out soon.
-- Dave
------------------------------
From: [EMAIL PROTECTED] (Phil Howard)
Subject: Re: Switchbox + mouse = braindead mouse?
Date: Thu, 08 Apr 1999 23:06:08 GMT
On Thu, 08 Apr 1999 16:55:38 GMT [EMAIL PROTECTED] wrote:
| I have a belkin switchbox connected up to a
| linux box. Occasionally (rare) when switching
| to the linux box, my mouse goes brain dead.
|
| The *only* solution to recover from this state
| is to reboot the box.
If gpm has the mouse open, just kill and restart gpm.
| Having written MS-mouse drivers before *and*
| looking at the kernel source, I think I know
| what the problem is.
|
| I hesitate to make any changes to the kernel
| unless I can (reliably) reproduce the problem
| to test. As stated, this only occurs "rarely"
| on my box and I cannot reproduce it "on demand"
|
| 1. Has anyone else experienced this problem?
| 2. Does anyone know how to reliably reproduce
| the problem?
I get this problem at times, with and without a box. The times
without a box are when I get up from my chair, which is a vinyl
covered pad kind of thing, and get a static charge. It discharges
into the frame of the chair and poses no danger to the computer
(as long as I make sure I discharge it only into the chair) but
many times, the discharge does seem to get into the long mouse
cord. What appears to be happening is that the bit shifter in
the serial connection via my AUX/PS2 port gets into some weird
state due to the spurious bits. Subsequent moving of the mouse
does not flush out this state, so it's not entirely clear that
is is merely a single extra bit. However, killing and restarting
gpm does fix it. It may be that gpm is hosed.
I do have X getting its mouse movements via gpm, so when I fix
gpm, it fixes X. If you have them sharing the device, you may
have to restart X to fix it, so I'd recommend having X get mouse
movements from gpm if indeed restarting gpm fixes your case.
| If you know how to reliably reproduce the problem,
| please post your reply AND email it to me.
I don't know that what I do is exactly your problem or not.
--
Phil Howard KA9WGN
[EMAIL PROTECTED] [EMAIL PROTECTED]
------------------------------
From: [EMAIL PROTECTED] (Kaz Kylheku)
Crossposted-To: comp.programming.threads
Subject: Re: gdb + linuxthreads + kernel 2.2.x = fixed :)
Date: Thu, 08 Apr 1999 23:08:31 GMT
On Thu, 08 Apr 1999 19:55:24 GMT, Paul Archard <[EMAIL PROTECTED]> wrote:
>Doing a manual load on the library using "sharedlibrary libpthread" solves
>the problem. Threads are then detected and debuggable (?!).
>
>Does anyone know if this behavior is "by design" or "by accident" ?
It's probably by design. The GDB patch adds LinuxThreads debugging ability by
making GDB peek at internal LinuxThreads data structures. Indeed, LinuxThreads
itself had to be modified to allow the hack to work by providing some extra
debugging info.
Presumably, without the symbols, GDB can't find the addresses of LinuxThreads
objects that it needs to access.
------------------------------
From: [EMAIL PROTECTED] (Phil Howard)
Subject: Re: 4 Gb memory? ECC?
Date: Thu, 08 Apr 1999 23:24:02 GMT
On 5 Apr 1999 02:30:57 GMT Andy Isaacson ([EMAIL PROTECTED]) wrote:
| AFAIK the max memory on Alpha is 2GB right now (2.2 with patches from
| axp-list). It's difficult to go higher than 4GB physical memory,
| because PCI devices can only address 4GB memory and the current
| architecture wants PCI devices to be able to DMA to anywhere in
| memory. The UltraSPARC port supports more than 4GB I think (I haven't
| checked but others claim that it does) but it does it by using bounce
| buffers.
I wonder if it might be possible to make memory above 4GB turn into
swap space. That would certainly be faster to swap in and out of than
virtually any bus device. It would still limit processes and buffers
to the low 4GB.
--
Phil Howard KA9WGN
[EMAIL PROTECTED] [EMAIL PROTECTED]
------------------------------
Date: Thu, 08 Apr 1999 17:34:15 -0700
From: "D. Stimits" <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Subject: Boot feature extension?
Hi:
I seem to have something of a catch-22 situation that SVGATextMode won't fix. My
particular ATI
video card has all the usual text modes, but initially the ones I want are not shown
during boot,
either by LILO or loadlin. A partial solution is to set it for vga=ask; however, this
always
requires me to wait for the menu and answer it during boots. Upon reaching the "ask"
screen, the
known modes are displayed, and there is also the option to enter "scan" as the mode,
and then it
will probe the card for further modes. This is where the modes are I want, in this
case "133" can be
entered.
Because I've memorized 133, I don't have to "scan" each time, I can just enter 133.
It'd seem
logical that 133 could be used as a boot parameter directly, in the loadlin command
line or lilo
append, but it can't. The catch 22 is that you must use a valid mode or it is
rejected, but the mode
isn't valid till you've scanned, which in turn requires you to be present. Because
this seems to be
both a LILO and a loadlin problem, and because kernel compiles attempting to compile
in this mode
are essentially just passing this value to the same interface, there appears to be no
way to
automate booting to this 133. I must do vga ask, and then sit there and type in the
answer.
SVGATextMode would at first seem to be the answer, but as it turns out, unless my text
mode columns
are as wide as the SVGATextMode to start with, it scrambles the first 2 letters of the
column across
the screen. If I've booted to the 132x50 mode I want, via ask and entering 133, I can
use
SVGATextMode to get a comparable mode. But I have to manually sit and enter 133 first
to get to a
mode SVGATextMode can handle and not scramble (it should produce the same results to
run
SVGATextMode, regardless of starting console geometry, but it doesn't). This is likely
a bug in
SVGATextMode, but even if it was not bugged, requiring this to boot to a valid mode
seems like a
workaround to a real fix.
What I'd like to see is the ability to append an override to LILO or loadlin, based on
kernel code
(and thus be able to compile this in to the kernel as default video mode), which would
allow one to
force what at first appears to be an invalid mode. If testing is desired, and it might
be a safety
item to consider, perhaps a command line switch could tell it to force scanning
silently, and then
to determine if the mode requested is valid. append="vga=scanfirst,133".
Thanks,
D. Stimits, [EMAIL PROTECTED]
------------------------------
From: Vipin Malik <[EMAIL PROTECTED]>
Subject: Re: Interrupts are ok, but ports in serial.c?
Date: Thu, 08 Apr 1999 18:40:00 -0500
If you do not need the serial support then just do not compile it into the
kernel at all.
Or best, you could make it into a module and load it as desired.
Also see the reply posted by another person by using setserial uart none,
I don't know if that will work, try it :)
Frank Haverkamp wrote:
> Hi Vipin,
>
> Thanks for your answer. In case of the interrupts your are right.
> I checked /proc/interrupts and /proc/ioports on my system with
> kernel 2.2.5 and must realise that the ports are requested by the
> serial driver. I put some debug information into serial.c. I
> replaced request_irq, free_irq, request_region and release_region
> by my own functions with added debug printout. Then I tried to
> cat a file to /dev/ttyS1 and the kernel messages said
>
> serial_request_irq(....)
> serial_free_irq(...)
>
> but nothing about request_region or release_region. The only messages
> about this I got during the bootup phase. I this a bug in serial.c and
> should I try to fix it or is there another way.
>
> Frank Haverkamp
>
> Vipin Malik wrote:
>
> > The serial driver should NOT be catching the interrupts and ports on
> > bootup UNLESS some device is opening them.
> > This is one of the features of the serial drivers (and the philosophy
> > behind most good drivers that use interrupts). They do not request the
> > interrupt from the kernel until actually needed.
> >
> > do a "cat /proc/interrupts"
> > If you see "serial" against int3 or 4 then some task has opened the
> > device (/dev/ttyS0 or 1 (or cua0 or 1)).
>
> --
> *** Frank Haverkamp
> *** [EMAIL PROTECTED]
> *** http://www.cs.tu-bs.de/~haver
> *** TU-Braunschweig (IBR)
------------------------------
Date: Thu, 08 Apr 1999 17:39:38 -0700
From: "D. Stimits" <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Subject: Ranish Partition manager
Hi:
I like what I've seen on Ranish partition manager, but it seems to lack information
concerning the
use of scsi with it. Does anyone here have opinions on whether this works (well) with
scsi drives?
2.1.x kernels?
Thanks,
D. Stimits, [EMAIL PROTECTED]
------------------------------
From: Vipin Malik <[EMAIL PROTECTED]>
Subject: Where can I get the e-mail for person responsible for ramdisk.c in kernel ?
Date: Thu, 08 Apr 1999 18:42:13 -0500
I need to mail the person responsible for maintaining rd.c (the ramdisk
driver) in the kernel and
the source code does not state his/her e-mail address. What's the
recourse ?
thanks
Vipin
------------------------------
From: [EMAIL PROTECTED] (Daniel Robert Franklin)
Subject: Re: Where can I get the e-mail for person responsible for ramdisk.c in kernel
?
Date: 8 Apr 99 23:50:46 GMT
Vipin Malik <[EMAIL PROTECTED]> writes:
>I need to mail the person responsible for maintaining rd.c (the ramdisk
>driver) in the kernel and
>the source code does not state his/her e-mail address. What's the
>recourse ?
Take a look in /usr/src/linux/CREDITS, look for ramdisk, you will find:
N: Theodore Ts'o
E: [EMAIL PROTECTED]
D: Random Linux hacker
D: Maintainer of tsx-11.mit.edu ftp archive
D: Maintainer of c.o.l.* Usenet<->mail gateway
D: Author of serial driver
D: Author of the new e2fsck
D: Author of job control and system call restart code
D: Author of ramdisk device driver
D: Author of loopback device driver
S: MIT Room E40-343
S: 1 Amherst Street
S: Cambridge, Massachusetts 02139
S: USA
- Daniel
--
******************************************************************************
* Daniel Franklin - Postgraduate student in Electrical Engineering
* [EMAIL PROTECTED]
******************************************************************************
------------------------------
From: [EMAIL PROTECTED] (Phil Howard)
Subject: Re: meminfo and ps
Date: Thu, 08 Apr 1999 23:49:37 GMT
On 02 Apr 1999 22:08:20 -0500 Michael Hirsch ([EMAIL PROTECTED]) wrote:
| Every time linux reads from the disk, it keeps what it read in
| memory. It doesn't wipe it out until that memory is needed, but then
| it is just a question of overwriting it--no caching to disk needed.
| The has the effect that memory gets filled up, but mostly with
| disposable stuff.
|
| I remember never using swap when I have 32 meg, but noticing that
| memory was tight. So I added another 64 meg, and filled it all up!
| Of course, I later realized it was almost all cached.
It would be nice if Linux could show such space under a notation called
perhapes "freeable". Any space that can be simply taken might fall
under this. But there can be different kinds of this space, too. Some
might be easier to take (read cache that isn't mmapped) than other space
(unmodified shared library code).
What I find to be a real problem in Linux is the WRITE data. Linux defers
writing to disk until the memory is full. This means that LOTS of memory
is used that cannot simply be taken freely for other purposes. I have a
machine set up with 256MB of RAM. I ran dd to zero out an 11.5GB disk so
I could start fresh with it, and it filled most of RAM up with some 200MB
of zeroes before the disk drive light even came on. This remains the one
major problem I see in the Linux kernel. There needs to be a limit on
this, settable per-fd, per-process, and per-system. The machine becomes
virtually unusable until all the I/O is done.
--
Phil Howard KA9WGN
[EMAIL PROTECTED] [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 (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
******************************
