Linux-Development-Sys Digest #164, Volume #7 Tue, 7 Sep 99 05:14:10 EDT
Contents:
Re: EROS, persistency, ext3fs? (David T. Blake)
Re: libc or glibc? (david parsons)
Re: Max threads and TCP connections? (Kaz Kylheku)
Re: GGI vs. framebuffer (david parsons)
Re: Intel740,MyCom drivers? ("Frank")
Re: GGI vs. framebuffer (Kaz Kylheku)
Re: Max threads and TCP connections? (Kaz Kylheku)
Re: Max threads and TCP connections? ("Andrey Fisunenko")
Re: Max threads and TCP connections? (Dan McGuirk)
Re: Can I install Linux on an IBM PS/2 model 95 XP 486? (=?iso-8859-1?Q?J=FCrgen?=
v.Hagen)
Re: TAO: the ultimate OS (EdToy)
Re: TAO: the ultimate OS (EdToy)
embedded linux ("Dr. Sven Bode")
Re: Counting hardware Interrupt ("Quiney, Philip (EXCHANGE:HAL02:HM10)")
Re: Linux standards compliance (david parsons)
No process cleanup after a core in 2.2.9 ("Markus M. Mueller")
----------------------------------------------------------------------------
From: [EMAIL PROTECTED] (David T. Blake)
Subject: Re: EROS, persistency, ext3fs?
Date: 7 Sep 1999 02:52:37 GMT
Reply-To: [EMAIL PROTECTED]
Peter Samuelson <[EMAIL PROTECTED]> wrote:
> ... Journalled filesystems are being developed for Linux (sct
> was working on one, don't know where he's at at this point) and are
> generally considered a Good Idea, except that they're (inherently?)
> somewhat slow for reading, which in many cases represents a lot more
> filesystem activity than writing.
Reiserfs journalling is being released as beta at the Kongress,
and ext3 with journalling is also being released. SCT is being
low key, and the reiserfs team is shouting it from the rooftops.
AFAIK, xfs is still a bunch of half-finished code and press
releases. But it really looks like 2.5 will bring THREE NEW
JOURNALLED filesystems to linux.
--
Dave Blake
[EMAIL PROTECTED]
------------------------------
From: o r c @ p e l l . p o r t l a n d . o r . u s (david parsons)
Subject: Re: libc or glibc?
Date: 6 Sep 1999 16:17:22 -0700
In article <[EMAIL PROTECTED]>,
Arthur Chiu <[EMAIL PROTECTED]> wrote:
>I'm new to Linux.
>
>People are talking about changing from libc to glibc. I'm using RH5.2.
>I can find descriptions of libc by 'info libc' but get nothing with
>'info glibc'. What's the difference between libc and glibc?
If you're using redhat 5.2, you're already using glibc.
Libc, in the libc vs. glibc context, is the old Linux C runtime
library. It used to have a stable interface years ago, but when the
ELF executable format was introduced, libc stability was tossed right
out the window giving everyone a wild "which application will the
libc of the week break today?" ride, and thus when the FSF restarted
their dormant libc, people were willing to ignore tiny details like
machine-independent machine dependent includes and a wholescale
scrapping of the existing binary interfaces so they could jump ship
from the existing library onto one that might actually work.
Glibc is (aggressively) buzzward compliant, so may be more
portable[1]. Glibc seems to be the preferred library for the
libstdc++ on Linux development efforts, so if you use C++ on Linux,
you'll save yourself a lot of hair-pulling if you join the hivemind.
There's some small support and development still going on for libc,
but that's support for libc4 in a non-politically correct executable
format. There's a lot of development going on for glibc, and
every distribution except one uses it as their preferred libraries,
so if writing portable code isn't your thing using glibc will give
you a lot more accidental portability than using the old libc.
Aaaaaand the Linux killer apps (civilization) use glibc, so you'll
need to have it on your system anyways.
____
david parsons \bi/ The latter is the reason I've got gl*bc loitering
\/ around on my development systems.
------------------------------
From: [EMAIL PROTECTED] (Kaz Kylheku)
Subject: Re: Max threads and TCP connections?
Reply-To: [EMAIL PROTECTED]
Date: Tue, 07 Sep 1999 04:32:50 GMT
On 6 Sep 1999 21:01:55 -0500, Peter Samuelson <[EMAIL PROTECTED]> wrote:
>[Bill LN <[EMAIL PROTECTED]>]
>> Is it possible to maintain 1 million TCP connections from a single
>> Linux box? What would the bandwidth and memory and OS ramifications
>> be if I try to hold 1 million TCP connections open at once ?
>
>TCP/IP numbers its ports with a 16-bit number, so there's a limit of
>65,536 ports per protocol (TCP or UDP). Each client connection
>requires one port, though server connections can share. (Presumably
>you're talking server connections so you'd be OK there.) In any case
>Linux is not designed to scale up to even close to 1 million
>connections or threads.
TCP connections are identified by four numbers. The source and destination IP
and the source and destination port number. That gives you a vast number of
different distinct connections, even if you constrain one endpoint.
In the case of a passive server, all incoming connections share the same
*local* port number and IP. For example, in the case of a HTTP daemon listening
on port 80, all browser connections have port 80 on the server side, and the
IP address of the network interface on which they are arriving. In principle,
there could be much more than 65536 such connections, because port numbers
are not a problem. Of course, there couldn't be more than 65536 from any
*single* IP.
So running out of port numbers should not be an issue in the case of *incoming*
connections, which is probably what you care about here.
Of course, any given host can only have so many *outgoing* connections, because
normally, each active connection is given a unique ephemeral local port number.
(This restriction could probably be overcome with some deep hackery.)
>One other thing to consider is where these connections are coming from.
>If they're all on the LAN, you start running into trouble with your
>Ethernet switches keeping track of a hundred thousand MAC addresses.
>If they're coming in from routers, keep in mind the hard TCP/IP limit
>of 65,536 connections between your server and each router. So if it's
What hard limit is that? Routers don't normally care about your connections,
they just forward IP packets. You don't connect to the router, you forward
connectionless datagrams to it. (But anything exotic that is sensitive to port
numbers might experience issues. For example, dynamic network address
translation or IP masquerading, that type of thing. And of course proxies---if
you used a proxy instead of letting clients connect to the server directly, the
proxy would quickly run out of port numbers. You probably wouldn't have these
things in your way if you wanted to support a huge number of connections.)
Note I'm not saying that you can have a million connections easily, only that
the limitations aren't going to be port-number related.
------------------------------
From: o r c @ p e l l . p o r t l a n d . o r . u s (david parsons)
Subject: Re: GGI vs. framebuffer
Date: 6 Sep 1999 16:26:18 -0700
In article <[EMAIL PROTECTED]>,
Nelson Minar <[EMAIL PROTECTED]> wrote:
>[EMAIL PROTECTED] (Christopher Browne) writes:
>> I don't think anybody much cares whether the graphical support for
>> consoles is accelerated; any hardware more modern than an 8MHz 68000
>> will provide quite ample performance for the support of text consoles.
>
>In general, I'd agree with you. But I had a Sparcstation 10 on my desk
>4 years ago that would lose several seconds a day from its clock. I
>used to log in on the text console to check email and stuff.
>
>Rumour had it the problem was the text console implementation blocked
>interrupts
Not all the time. Just when it scrolls. It's possible to severely
confuse a Sparc if you boot up a text console, log in, and set a
brick on the return key.
____
david parsons \bi/ Or try scrolling through a 1,000,000 line logfile.
\/ You'll grow old and die before you reach the bottom
of the file, so that's a fairly moot point.
------------------------------
From: "Frank" <[EMAIL PROTECTED]>
Subject: Re: Intel740,MyCom drivers?
Date: Tue, 7 Sep 1999 13:23:49 -0400
I am new to linux, but as for Intel740, at least I found it in the Mandrake
6.0 CD-ROM, under /apps. You can download the whole CD at ftp.cdrom.com, or
other link found at www.mandrake.com.
Hope this help.
JBFrandsen wrote in message ...
>Hi!
>Are there anybody who knows were to find some drivers to my Intel740 screen
>card, or to my MyCom screen?
>
>
>Sorry, about my bad english, I am from Denmark !!!
>
>
>/Jack Frandsen
>
>
------------------------------
From: [EMAIL PROTECTED] (Kaz Kylheku)
Subject: Re: GGI vs. framebuffer
Reply-To: [EMAIL PROTECTED]
Date: Tue, 07 Sep 1999 05:41:12 GMT
On 6 Sep 1999 16:26:18 -0700, david parsons <[EMAIL PROTECTED]> wrote:
> Not all the time. Just when it scrolls. It's possible to severely
> confuse a Sparc if you boot up a text console, log in, and set a
> brick on the return key.
Repeat following to customer and observe resulting facial expressions: ``The
system you are using places a higher priority on scrolling its screen than on
processing transactions. In fact, scrolling the screen is so important
that no other processing can take place while it happens.''
------------------------------
From: [EMAIL PROTECTED] (Kaz Kylheku)
Subject: Re: Max threads and TCP connections?
Reply-To: [EMAIL PROTECTED]
Date: Tue, 07 Sep 1999 05:42:17 GMT
On Tue, 07 Sep 1999 04:32:50 GMT, Kaz Kylheku <[EMAIL PROTECTED]> wrote:
>there could be much more than 65536 such connections, because port numbers
>are not a problem. Of course, there couldn't be more than 65536 from any
>*single* IP.
Actually, is zero a valid port number? The sockets interface won't let
you bind to zero since it means ``choose an arbitrary port for me''.
------------------------------
From: "Andrey Fisunenko" <[EMAIL PROTECTED]>
Subject: Re: Max threads and TCP connections?
Date: Tue, 7 Sep 1999 09:21:30 +0300
>
>Bill LN wrote:
>>
>> Is it possible to maintain 1 million TCP connections from a single
>> Linux box? What would the bandwidth and memory and OS ramifications be
>> if I try to hold 1 million TCP connections open at once ?
>
> In my experience, 16,000 seems to be the maximum. I encountered some
>nasty memory allocation problems in the Linux kernel above that.
>
We try to run sleeping threads by call pthread_create (on Red Hat with
kernel 2.2.5)
and increase limitations by set-limits.
But we can't get more then about 400 ptrheads.
What's wrong?
How you got 16000? Those test compiled and executed under FreeBSD creates
pthreads while system has enough memory.
It seems to be one of the weakest sides of Linux?
And all tend to reply in book manner -
"You don't need that amount of thread".
But "if stars light in night someone needs that".
_There are_ tons of tasks where threads are more natural than fork or
something like that.
Some tasks can not be solved without threads at all.
------------------------------
From: Dan McGuirk <[EMAIL PROTECTED]>
Subject: Re: Max threads and TCP connections?
Date: Mon, 06 Sep 1999 23:35:48 -0700
Bill LN wrote:
> Is it possible to maintain 1 million TCP connections from a single
> Linux box? What would the bandwidth and memory and OS ramifications be
> if I try to hold 1 million TCP connections open at once ?
Well, there's a certain amount of memory that's used just for accounting
purposes for each connection. For example, each connection will have a
'struct sock' allocated for it in the kernel. IIRC, this structure is
about 200 bytes, so for a million connections you're using 200 megs of
memory right there. Each connection will also have both read and write
socket buffers; in Linux these are not allocated to a fixed size, but
can expand and contract, up to a maximum of 64K in each direction by
default. If you have any connections going over high-speed links with
large latencies, or you start getting packet loss, or your application
doesn't process incoming data quickly, you'll start using up space in
these buffers; this can be up to 128K per connection (128 gigs for a
million connections!).
To find the socket an incoming TCP packet goes to, Linux uses a hash
table that by default contains a maximum of 512 buckets. If you have a
million connections, you're going to have to scan through a linked list
of about 2000 entries for each packet, which is going to be a huge
performance hit.
If you really want to have this many people talking to you at the same
time, you should probably use a non-connection-oriented protocol (UDP),
and figure out how to deal with packet losses in your application.
--
Dan McGuirk
[EMAIL PROTECTED]
------------------------------
From: =?iso-8859-1?Q?J=FCrgen?= v.Hagen
Subject: Re: Can I install Linux on an IBM PS/2 model 95 XP 486?
Date: Tue, 07 Sep 1999 08:16:53 +0200
Matan Ziv-Av schrieb:
>
> On Mon, 6 Sep 1999, [iso-8859-1] J�rgen v.Hagen wrote:
>
> > Alan Smeltzer schrieb:
> > >
> > > I have an IBM PS/2 model 95 XP 486.This thing uses an IBM SCSI controller
> >
> > as I've written in another posting, it wasn't difficult to
> > do this on my 77.
> >
> > However I have had no luck to use svgalib on the MCA machine.
> > It complains about PCI type 1 not being supported.
> >
> > Has anybody managed to get sgvgalib running on an MCA?
>
> This is a problem with the et6000 driver , that was fixed in 1.4.0, but it
> should not prevent svgalib from working. Does the PS/2 machine have a vga
> compatible video card?
I have different informations about the card:
XGA
Enhanced Localbus Graphis (that's on a DOS diskette for it)
and the Windows GRaphics Setup tells me, that it is a VESA
I guess I should have a look at the physical card to see what
it says.
WRT the error message of svgalib, you're right that it
does work after this message, however accesses to the
ZIP on the parallel port fail afterwards.
juergen
------------------------------
From: [EMAIL PROTECTED] (EdToy)
Crossposted-To: alt.os.linux,comp.os.linux.advocacy,comp.os.misc,comp.unix.advocacy
Subject: Re: TAO: the ultimate OS
Date: Tue, 7 Sep 1999 00:43:12 -0500
In article <7r0m7k$[EMAIL PROTECTED]>, [EMAIL PROTECTED]
says...
> In article <[EMAIL PROTECTED]>,
> EdToy <[EMAIL PROTECTED]> wrote:
> > those who
> >do the work, don't get the rewards of it.
>
> Those who do the work on Linux get the rewards - namely, a free, powerful
> and stable operating system.
>
> There are more rewards in life than money.
Well if it was a hobby for those who did it, fine. I tend to think of
programming as work, not that it's not fun sometimes, but still work.
Ed
------------------------------
From: [EMAIL PROTECTED] (EdToy)
Crossposted-To: alt.os.linux,comp.os.linux.advocacy,comp.os.misc,comp.unix.advocacy
Subject: Re: TAO: the ultimate OS
Date: Tue, 7 Sep 1999 01:07:05 -0500
In article <7qvj4a$[EMAIL PROTECTED]>,
[EMAIL PROTECTED] says...
> note above I did not declare anything "true by fiat". I proposed
> an alternative to existing dogma & conventional wisdom. I am
> saying, "imagine a universe exists in which you are not confined to
> the boxes you now believe you are confined within"<g>
Well don't let your head swell up too much because almost any comp sci
student could have come up with (and they probably have) with your subset
of suggestions (and many others too). You may think that you're thinking
"outside of the box" but indeed you've simply followed obvious paths of
the "existing dogma & conventional wisdom". You may have done some
_extrapolation_, but I'd have to read your ideas again to be sure. If
you want to "see" outside of the box, you'll have to step back a "wee"
bit further. Again, I believe the proof is in the doing, not in the
dreaming, and using that experience to _develop_ your preliminary ideas
into a vision. Not that dreaming doesn't have any merit, just that it's
on par with masturbation.
Ed
------------------------------
From: "Dr. Sven Bode" <[EMAIL PROTECTED]>
Subject: embedded linux
Date: Tue, 07 Sep 1999 09:18:54 +0200
is there a xwindows linux system which will work with less then 10 MB
RAM and about 16MB Flashdisk.
If so, what development tools for application is recommended ?
Any help from you will help us to get rid of M$ ;-)
--
Sven Bode
[EMAIL PROTECTED]
Tel ++49 30 69 0003 13
------------------------------
From: "Quiney, Philip (EXCHANGE:HAL02:HM10)" <[EMAIL PROTECTED]>
Subject: Re: Counting hardware Interrupt
Date: Tue, 07 Sep 1999 08:05:55 +0100
Simon Kwan wrote:
>
> Hi,
> I want to count the number of electrical pulses coming into the CPU IRQ
> line. The count need to increment throughout the life of the system. Hence,
> may be I can write the 'number of pulses' to a disk file as soon as it come
> it. So, when the machine was power down and re-boot some on the next day, it
> will read from disk the last count and continue.
What sort of rate are you expecting. This would cause disk thrashing if
the interrupts are coming in quickly.
A driver can create a file under /proc which is a virtual file. The
contents of this could be your interrupt count. When the machine is shut
down a script could be triggered (via /etc/rc.d/init.d) to save the
count to a disk file (ie cat /proc/irq_count > /var/log/irq_count). The
saved count could then be passed to the driver when it is loaded on boot
up. Remember at some point during shutdown the file system is unmounted
and becomes 'unavailable'.
>
> Do I need to go the the length of learning how to write a Linux device
> driver in order to perform the above task? Understand that the device
> driver is allowed to interface with hardware interrupt line handling. Can it
> write to file?
You wil have to write some code to handle the interrupt. In your case
increment a suitable variable and then whatever the device requires to
make it clear the interrupt condition - otherwise you will have a
permanent interrupt to deal with and your linux kernel will appear to
have locked up.
Remember that the variable even if an 'unsigned long' will eventually
roll over back to 0. The time it takes to do this will depend on the
frequency of the interrupt.
Get hold of a copy of "Linux Device Drivers" by Alessandro Rubini. There
are some simple code examples in there which should help.
Regards
Phil Q
--
Phil Quiney Digital PowerLine,
[EMAIL PROTECTED] Nortel Networks,
Telephone: +44 (1279) 402363 London Rd, Harlow,
Fax: +44 (1279) 402885 Essex CM17 9NA,
United Kingdom.
"This message may contain information proprietary to Northern
Telecom so any unauthorised disclosure, copying or distribution
of its contents is strictly prohibited."
------------------------------
From: o r c @ p e l l . p o r t l a n d . o r . u s (david parsons)
Subject: Re: Linux standards compliance
Date: 6 Sep 1999 21:20:03 -0700
In article <[EMAIL PROTECTED]>,
Nix <$}xinix{[email protected]> wrote:
>[EMAIL PROTECTED] (Christopher Browne) writes:
>
>> It's not evident that much actual work has been put into rewriting
>> Linux device drivers to conform with the UDI API
>
>AFAIK, none at all has.
>
>I doubt it will happen at all (unless the kernel forks) while the Powers
>that Be in the Linux kernel world are so against it. (Oh dear, *what* a
>pity. ;} )
It is a pity; currently one of the sad side-effects of upgrading to
newer kernels is that I get to discard Even More Hardware that the
latest and greatest device drivers won't handle (I have, as exhibit
(a), the NCR810 card that I had to pull out of my development
machine because the 2.3.15+ device drivers won't recognise it
anymore.) And don't forget the filesystems that aren't supported
anymore because nobody is willing to be a maintenance programmer
and change the kernel hooks every six months for the rest of all
time, because the Precious Bodily Fluids of the kernel would be
tainted by publishing a driver interface and sticking with it.
____
david parsons \bi/ Now to print out all the .pdfs and see how quickly
\/ a reference implementation can be hacked out.
------------------------------
From: "Markus M. Mueller" <[EMAIL PROTECTED]>
Subject: No process cleanup after a core in 2.2.9
Date: Tue, 07 Sep 1999 09:17:21 +0200
I'm running Mandrake 6.0 with a 2.2.9 kernel.
ulimit -c unlimited
When a program core dumps no core file is written ( file is there size
is zero ), and that process HANGS!!! Also kill -9 has no effect only a
reboot removes it from the process tree.
Anybody the same behaviour?
Markus
------------------------------
** 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
******************************
