Very Computer

Hello,

I'm running a dual P2 - 233 MHZ processor Linux server with Slackware 3.4
(Kernel 2.0.33 complied for SMP).

Here are some specs about the server
2 -  233 MHZ Pentium 2 processors
Asus P2L97-DS Motherboard (USB and Power saving features turned off)
Using Scsi adapter on the board controlling 2 Seagate Barracuda ULTRA WIDE
drives
128 MB of DRAM
3com 905 Network Card (0.49 drivers)
Generic Trident svga card

After a period of time the server just crashes, this can be after one day
or after 14 days. When it crashes the screen turns black and the server
shows no sign of live, the Keyboard does not respond neither can you ping
the server. There are no indications on why the server crashes... (no
entries in the logs !!!).
The Power lights on the machine are still on when it crashes and the drives
are still spinning. The only way to get this machine back to live is by
turning it's power off and back on. I called Asus to find out if there may
be a problem with the settings but everything seemed to be fine, so they
told me to replace the board which I did but the machine keeps crashing.
And yes I do have the following statement in the Lilo.conf file mem=127
(total size of memory - 1MB). I even got the newest scsi drivers for the on
board 2940 adaptec. I'm running out of ideas...
Should I try a different kernel, lets say 2.1.90 ?
Is there much improvement in terms of SMP in the 2.1.xx kernels ?
Does anybody think this is caused by the kernel ?

Help...

Thanks
Jorg B.

Ps. We are running about 11 "single processor" linux servers with the same
hardware , except the board is different - none of these servers ever
crashes...

regards Henrik

--
spammer strikeback:


Join LinuxNet RC5! Visit http://www.linuxnet.org for info

Hello!

Quote:>I'm running a dual P2 - 233 MHZ processor Linux server with Slackware 3.4
>(Kernel 2.0.33 complied for SMP).

>Here are some specs about the server
>2 -  233 MHZ Pentium 2 processors
>Asus P2L97-DS Motherboard (USB and Power saving features turned off)
>Using Scsi adapter on the board controlling 2 Seagate Barracuda ULTRA WIDE
>drives
>128 MB of DRAM
>3com 905 Network Card (0.49 drivers)
>Generic Trident svga card

>After a period of time the server just crashes, this can be after one day
>or after 14 days. When it crashes the screen turns black and the server

Ciao,
Mario.

I'll bet it's more likely a hardware conflict. Like Mario said, I'd
try swapping out the ethernet card...

Best regards,
--

Black holes are where God divided by zero.

On Sun, 22 Mar 1998 14:41:52 +0100, Mario Procopio wrote in alt.linux (too much anyway):

Quote:>Hello!

>>I'm running a dual P2 - 233 MHZ processor Linux server with Slackware 3.4
>>(Kernel 2.0.33 complied for SMP).

>>Here are some specs about the server
>>2 -  233 MHZ Pentium 2 processors
>>Asus P2L97-DS Motherboard (USB and Power saving features turned off)
>>Using Scsi adapter on the board controlling 2 Seagate Barracuda ULTRA WIDE
>>drives
>>128 MB of DRAM

Maybe this? Did you add 'append mem=127M' to your /etc/lilo.conf?
Some machines use 128Mb, some 128Mb-384Kb, some the other values. 127Mb
is safe, and you don't loose much.

And, 2.1.x kernels have better SMP support. Consider upgrading, or wait
for 2.2.0, it shouldn't be too long.

--

Zagreb, Ivanicgradska 48.  tel. 385 01 227760, job: 6150830.

-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GS d- s: a- C++ UL++++$ P+ L++$ E--- W+++ N++ o K- w--- O M- V-- P S+++
P E++ Y+ PGP+ t !5 X- !R tv--- b++ !DI D++ G++ >e+++ hr- y+ z+
------END GEEK CODE BLOCK------

"Bugs are not going to inherit the earth. They own it now. So we might as
   well make peace with the landlord." -- T. Eisner (1989)

Put plainly, the network preformance sucks.  The hub has four LEDs (1%,
10%, 20%, >30%) for utilization, and I *never* get above 20%.  I've done
testing of transfers over ftp, SMB, etc.  I have also measured the
performance using a command-line Win32 app for reading files on the SMB
shares and done some tests using ftp.  3MB/s is about the max I have
seen.

Disk performance is not a problem as I am reading off of a 6-disk RAID0
array that can saturate the FW SCSI bus.  The disk on the Win95 box can
also spit out over 6MB/s sustained.  

One thing I have noticed is that ftp transfers to the Win95 from the
Linux box are *much* faster than the opposite.  (doing a 'put' from the
Win95 machine is painful)  The hub shows a burst of activity, then about
a second pause, then another burst, etc.  The transfer is sustained
while doing a 'get' on the Win95 machine.  

The SMP box is a dual PPro/233; the Win95 is a K6/233.  Now the question
is which box is the dog on net performance?  I'm running RH5 upgraded to
2.0.33.  My NT boxes at work have no problems cranking out over 7MB/s
(limited by the HD speed in the simple test I did).  My Win95 machines
are all on 10Mb, so I don't know whether Win95 can handle the load or
not.  

Thanks for any help.  

-Wayne

  >> > 3com 905 Network Card (0.49 drivers)

  >> This is a guess, but could you try to replace the  network card with
  >> another one? I have some bad experience of 3c900 myself which uses the
  >> same driver. I have seen machines freeze, but it doesn't happen that
  >> often and only at heavy netload, for example when doing rdist to to
  >> other machines at the same time. I'm not sure if my problem is because
  >> of the network card, but I think so.

  Wayne> I just put Linux back on my SMP box after I got fed up with NT.  I have
  Wayne> a second machine running Win95 (for MSOffice, etc).  The two are
  Wayne> connected via a 4-port NETGEAR FE104 100Mb hub.  The machines have 3C905
  Wayne> NICs.  I'm doing IP Masquerading (the Linux box has the modems -- EQL is
  Wayne> next), but my main problem is with the speed of the 100Mb link.  I
  Wayne> haven't had any crashes on Linux yet, but I have not stressed the system
  Wayne> (aside from having the CPUs cracking rc5).

  Wayne> Put plainly, the network preformance sucks.  The hub has four LEDs (1%,
  Wayne> 10%, 20%, >30%) for utilization, and I *never* get above 20%.  I've done
  Wayne> testing of transfers over ftp, SMB, etc.  I have also measured the
  Wayne> performance using a command-line Win32 app for reading files on the SMB
  Wayne> shares and done some tests using ftp.  3MB/s is about the max I have
  Wayne> seen.

  Wayne> Disk performance is not a problem as I am reading off of a 6-disk RAID0
  Wayne> array that can saturate the FW SCSI bus.  The disk on the Win95 box can
  Wayne> also spit out over 6MB/s sustained.  

  Wayne> One thing I have noticed is that ftp transfers to the Win95 from the
  Wayne> Linux box are *much* faster than the opposite.  (doing a 'put' from the
  Wayne> Win95 machine is painful)  The hub shows a burst of activity, then about
  Wayne> a second pause, then another burst, etc.  The transfer is sustained
  Wayne> while doing a 'get' on the Win95 machine.  

  Wayne> The SMP box is a dual PPro/233; the Win95 is a K6/233.  Now the question
  Wayne> is which box is the dog on net performance?  I'm running RH5 upgraded to
  Wayne> 2.0.33.  My NT boxes at work have no problems cranking out over 7MB/s
  Wayne> (limited by the HD speed in the simple test I did).  My Win95 machines
  Wayne> are all on 10Mb, so I don't know whether Win95 can handle the load or
  Wayne> not.  

  Wayne> Thanks for any help.  

  Wayne> -Wayne

The attributions in the above don't indicat who said

  >> > 3com 905 Network Card (0.49 drivers)

but my 3c905 performance dramatically improved when I upgraded to that
version; previously I had similar problems to those you describe, but
from what I could see initially the problem appeared not to be the
Linux machine's fault since I could see from tcpdump that the Win95
box was pausing a long time between packets.

In fact this was not the case; the Linux box's 3c095 card/driver had
been dropping the packets so tcpdump never saw them...

Is there anyone here using 3Com NICs that can get 7+ MB/s on fast
ethernet with Linux?  

-Wayne

-Greg Mildenhall

Does anyone here get good performance on 100Mb ethernet with Linux?  

(that is, over 6 MB/s)

-Wayne

Quote:> I have 2 machines with this card installed (a p5/200 and a k6/200).  I put
> linux on both of them and copied a cdrom image (650meg or so) from hard disk
> to hard disk.  I think I got about 5.5MegaBytes/sec (via ncftp's report).

Quote:> when going between the same machines, using linux and w95, the copy was MUCH
> slower.  so slow, that I terminated the copy after 5 minutes or so - the
> ncftp progress meter showed so little progress, I didn't want to wait around
> ;-)  there must have been some timing/race conditions between the two
> implementations (drivers for the tulip chip).  (the reason I needed to copy
> this data was because I built an iso9660 image on linux, but I like to burn
> cd-r's on my wintel machine (the mastering software on wintel is better
> (sorry) than what I have for linux, and my writer is attached to the wintel
> box).

Quote:> anyway, consider the tulip cards.  they're fast, cheap, and the driver is
> pretty stable (under linux, at least).

-Wayne