samba - Jan 2003 - linux server completely hangs copying files with samba

hi,
i'm experiencying a strange problem with a mandrake 9.0 machine,
kernel 2.4.19, samba (different versions, including latest).
when I copy large amount of data from any win98 machine to a share 
directory on linux, after a random time the linux machine hangs; not 
responding even to ctrl-alt-canc, only possibility is to reset.
I was using a realtek 8139 (driver 8139too) which I know sometime giving 
problems; I tried a Dlink (always 8139too), same problem; reading around it 
could be a driver problem so I managed to have a good 3com905c (driver 
3c59x); big surprise: same problem!!!
i noticed that trasferring from the laptop which has a 10 Mit/s card 
doesn't give the error, so probably it happens with large bandwith usage on 
the server.
nothing appears on any log, as if nothing appened.
anybody experienced similar problems? especially anybody solved it?

regards

david

>i'm experiencying a strange problem with a mandrake 9.0
>machine, kernel 2.4.19, samba (different versions,
>including latest). when I copy large amount of data from
>any win98 machine to a share directory on linux, after a
>random time the linux machine hangs; not responding even
>to ctrl-alt-canc, only possibility is to reset. I was
>using a realtek 8139 (driver 8139too) which I know
>sometime giving problems; I tried a Dlink (always 8139too),
>same problem; reading around it could be a driver problem
>so I managed to have a good 3com905c (driver 3c59x); big
>surprise: same problem!!! i noticed that trasferring from
>the laptop which has a 10 Mit/s card doesn't give the
>error, so probably it happens with large bandwith usage
>on the server.
I think it's the same problem I had and it's not the
network problem. It's about a problem with system on IDE
drives. After I switched to a SCSI disk for boot, swap and
root the problem disappeared completely.

The fact that the problem doesn't occur with 10 Mbps
connections only shows that slow ether cannot create enough
stress for the problem to manifest itself. It's some
racing condition in UDMA driver. If you disable DMA on
your system IDE drive you won't have the problem but the
performance really sucks.



_____________________________________________________________
Get 25MB, POP3, Spam Filtering with LYCOS MAIL PLUS for $19.95/year.
http://login.mail.lycos.com/brandPage.shtml?pageId=plus&ref=lmtplus

>To: samba@lists.samba.org
>Date: Sat, 25 Jan 2003 11:10:43 +0100
>From: "Dragan Krnic" <dkrnic@lycos.com>
>Cc: davide.cavaleri@libero.it
>Reply-To: dkrnic@lycos.com
>Subject: [Samba] linux server completely hangs copying files with samba
>Organization: Lycos Mail  (http://www.mail.lycos.com:80)
>
>I think it's the same problem I had and it's not the
>network problem. It's about a problem with system on IDE
>drives. After I switched to a SCSI disk for boot, swap and
>root the problem disappeared completely.
>
>The fact that the problem doesn't occur with 10 Mbps
>connections only shows that slow ether cannot create enough
>stress for the problem to manifest itself. It's some
>racing condition in UDMA driver. If you disable DMA on
>your system IDE drive you won't have the problem but the
>performance really sucks.
Which IDE chipset? We ran our network on a machine runnning essentially
9.0 (cooker but with 2.4.19-16mdk kernel, and most server components and
libraries haven't changed) on a 120GB IDE disk for a week with no
problems, with dma enabled. So it might be chipset specific.

Maybe if you two can log a bugzilla or Mandrakeexpert entry, and you may
be able to try the kernel update that should be coming and see if it
addresses the problem.

Regards,
Buchan

-- 
|----------------Registered Linux User #182071-----------------|
Buchan Milne                Mechanical Engineer, Network Manager
Cellphone * Work            +27 82 472 2231 * +27 21 8828820x121
Stellenbosch Automotive Engineering         http://www.cae.co.za
GPG Key                   http://ranger.dnsalias.com/bgmilne.asc
1024D/60D204A7 2919 E232 5610 A038 87B1 72D6 AC92 BA50 60D2 04A7

>>I think it's the same problem I had and it's not the
>>network problem. It's about a problem with system on IDE
>>drives. After I switched to a SCSI disk for boot, swap and
>>root the problem disappeared completely.
>>
>>The fact that the problem doesn't occur with 10 Mbps
>>connections only shows that slow ether cannot create enough
>>stress for the problem to manifest itself. It's some
>>racing condition in UDMA driver. If you disable DMA on
>>your system IDE drive you won't have the problem but the
>>performance really sucks.
>
>Which IDE chipset? We ran our network on a machine runnning 
>essentially 9.0 (cooker but with 2.4.19-16mdk kernel, and most 
>server components and libraries haven't changed) on a 120GB
>IDE disk for a week with no problems, with dma enabled. So
>it might be chipset specific.
Good question! The chipset in my case was i845 from Intel.
I had it running for a month in a sandbox with only a couple
of clients, but when I eventually took it into production and
added 40 clients it would simply freeze from time to time 
necessitating a cold start.
>Maybe if you two can log a bugzilla or Mandrakeexpert entry,
>and you may be able to try the kernel update that should be
>coming and see if it addresses the problem.
My distro is SuSE 8.1, but 2.4.19 is 2.4.19 is 2.4.19, right?
I'd never go back to IDE even if I knew the problem is
solved. At my home I can use anything, it doesn't matter.
But a file server may be privileged with a couple of decent
SCSI drives if you're serious. And besides, 15 Krpm LVD disks
are a lot faster than 7200 IDEs and I can string many more on
the same wire.


_____________________________________________________________
Get 25MB, POP3, Spam Filtering with LYCOS MAIL PLUS for $19.95/year.
http://login.mail.lycos.com/brandPage.shtml?pageId=plus&ref=lmtplus

I have been using Samba for some time now and have had no problems. I now
have users that want to use XP and am not able to join them to the Samba
domain. I remember a while back that somebody had a registry update for XP
but I cannot locate it.

Would some kind soul please send it to me

Best regards
Stan

On Sat, 2003-01-25 at 15:35, Stan S. wrote:
> I have been using Samba for some time now and have had no problems. I now
> have users that want to use XP and am not able to join them to the Samba
> domain. I remember a while back that somebody had a registry update for XP
> but I cannot locate it.
> 
> Would some kind soul please send it to me
> 
google for signorseal

brad

Thanks again Brad! It works great. Problem Resolved.

Stan
----- Original Message -----
From: "Bradley W. Langhorst" <brad@langhorst.com>
To: "Stan S." <sgs9198@attbi.com>
Cc: <samba@lists.samba.org>
Sent: Saturday, January 25, 2003 12:58 PM
Subject: Re: [Samba] XP Registry Hack

> On Sat, 2003-01-25 at 15:35, Stan S. wrote:
> > I have been using Samba for some time now and have had no problems. I
now
> > have users that want to use XP and am not able to join them to the
Samba
> > domain. I remember a while back that somebody had a registry update
for
XP
> > but I cannot locate it.
> >
> > Would some kind soul please send it to me
> >
> google for signorseal
>
> brad
>
>

>> >Which IDE chipset? We ran our network on a machine runnning
>> >essentially 9.0 (cooker but with 2.4.19-16mdk kernel, and most
>> >server components and libraries haven't changed) on a 120GB
>> >IDE disk for a week with no problems, with dma enabled. So
>> >it might be chipset specific.
>>
>> Good question! The chipset in my case was i845 from Intel.
>> I had it running for a month in a sandbox with only a couple
>> of clients, but when I eventually took it into production and
>> added 40 clients it would simply freeze from time to time
>> necessitating a cold start.
>>
>
>IIRC, the i845 does give problems with DMA enabled under 2.4.19,
>I think smoe people are having more success with the latest 
>kernels ...
It is a sensitive thing when 40 other guys depend on it.
You don't have much time to fix bugs in kernel. Your
mission is to let the users work. So you cut corners.
>
>> My distro is SuSE 8.1, but 2.4.19 is 2.4.19 is 2.4.19, right?
>> I'd never go back to IDE even if I knew the problem is
>> solved. At my home I can use anything, it doesn't matter.
>> But a file server may be privileged with a couple of decent
>> SCSI drives if you're serious. And besides, 15 Krpm LVD disks
>> are a lot faster than 7200 IDEs and I can string many more on
>> the same wire.
>
>And even better if you cah do hardware RAID ... and then use LVM 
>over that ;-). Our new Dell PowerEdge 2500 has 5*36GB/15k LVDs
>setup like that, and is stable as a rock (so far, it's only been
>in production for a few weeks).
Yeah. I had to make do with a Compaq 5300 controller and 
StorageWorks enclosures with 72 GB disks. Of course LVM
is a must to slice the RAID into a boot, swap, root and
data volumes. So if the PC burns I can just plug the whole
box into another PC. It rocks a lot faster if you can
stripe-n-mirror but I had to fall back to RAID-5 because
a) the people needed more place than speed and b) it is
still far faster than a non-mirrorred stripe I made with
fast IDE drives. And besides, I switched to reiserfs -
lot of performance gains in directory manipulation
operations.


_____________________________________________________________
Get 25MB, POP3, Spam Filtering with LYCOS MAIL PLUS for $19.95/year.
http://login.mail.lycos.com/brandPage.shtml?pageId=plus&ref=lmtplus

so, after infinite tries and different combination of bios settings, 
modules loaded, ethernet adapter i concluded that what makes the system 
crash is having the ----  usb-ohci.o ----- module loaded; i removed all the 
modules installed by mandrake and it was ok for at least 100 GB of 
transfer; loading the usb-ohci.o module even during copying makes the 
system crash after a time between few seconds and 5 minutes. when it 
crashes the ide led remains lit, so must be some problem involved with ide 
as dragan said. making few changes in the bios i had ether card and usb 
sharing irq 9; i thought was the reason, but letting the ether card have 
irq 9 and the usb irq 11 was giving the same problem; pratically speaking i 
don't need usb, so i can leave the module unloaded or even better disable 
it from bios; but teoretically speaking it would be nice to understand what 
is going on; with usb-ohci.o loaded it happens only with large transfer 
rates, so disabling dma is not giving crash; even ricompiling the kernel, 
which i don't understand why the transfer rate of the ethernet card become 
slower (any suggestion why?) saves from hangs.
i think i'll sleep anyway and that's what i'm going to do.
regards

david

>so, after infinite tries and different combination of bios
>settings, modules loaded, ethernet adapter i concluded that
>what makes the system crash is having the ----  usb-ohci.o
>----- module loaded; i removed all the modules installed by
>mandrake and it was ok for at least 100 GB of transfer;
>loading the usb-ohci.o module even during copying makes the 
>system crash after a time between few seconds and 5 minutes.
>when it crashes the ide led remains lit, so must be some
>problem involved with ide as dragan said. making few changes
>in the bios i had ether card and usb sharing irq 9; i thought
>was the reason, but letting the ether card have irq 9 and the
>usb irq 11 was giving the same problem; pratically speaking i
>don't need usb, so i can leave the module unloaded or even
>better disable it from bios; but teoretically speaking it
>would be nice to understand what is going on; with usb-ohci.o
>loaded it happens only with large transfer rates, so
>disabling dma is not giving crash; even ricompiling the
>kernel, which i don't understand why the transfer rate of the 
>ethernet card become slower (any suggestion why?) saves from
>hangs.
There must be more to it than just usb-ohci.o. Disabling
USB from BIOS didn't do any good for me. On the other hand,
even after installing the system on a SCSI drive I used two
IDE drives in a striped configuration for the samba shares
and never had any problems with it.

In my experience the kernel panic only occurs if some system
partition, probably swap, resides on an IDE disk with DMA.
How USB module enters the picture is not clear to me. Both
USB and IDE are on-board. SCSI is not. A Promise PCI IDE card
would probably work without problem.


_____________________________________________________________
Get 25MB, POP3, Spam Filtering with LYCOS MAIL PLUS for $19.95/year.
http://login.mail.lycos.com/brandPage.shtml?pageId=plus&ref=lmtplus

>> final experiments;
>> after realizing that unloading the usb-ohci module (non even the
>> usbcore) system was fine, i compiled the kernel without modules
>> support and without usb. the machine crashed and actually i was 
>> really angry, becuase i thought i had found the problem; but 
>> finally i discovered that if usb is not loaded the machine 
>> crashes, but if i load it and unload it (going back to the same 
>> situation) it doesn't crash. obviously if i disale usb from bios 
>> everything in ok. maybe is something about irq, so if it is 
>> enabled in bios the ira (11 for me) is used. loading and
>> unloading the module will do something strange to the irq
>> handles. cat /proc/interrupts gives 0 on usb. at this point i'm 
>> curios to find out why this happens; since i've red of
>> many people having this problem maybe someone (not you, as you
>> told me) had a similar situation.
>> don't know if it's relevant, but only once the keyboard was 
>> blocked, i had to unplag and plug it back (ps/2 keyb) and it was 
>> ok, checking the interrupts they had jumped from 1500 to about 
>> 3000 in few seconds. going to work, nice day
>
>I forget how recent your hardware is, but have you tried booting with:
>
>lilo: linux noapic
That's one of the fírst things I tried. Didn't help, but nodma
(or something similar disabling IDE DMA) did.


_____________________________________________________________
Get 25MB, POP3, Spam Filtering with LYCOS MAIL PLUS for $19.95/year.
http://login.mail.lycos.com/brandPage.shtml?pageId=plus&ref=lmtplus