Very Computer

Hi there,

I've setup two servers with almost identical hardware, and identical
software. I've noticed a few problems which I haven't been able to
resolve.

Server #1:
AMD Athlon 750
Asus K7M motherboard
512 Megs of 3.3 volt ECC SDRAM
3COM 3c905 ethernet card
Slackware 7
Kernel 2.2.14
Seagate ULTRA WIDE SCSI 2 LVD 9 Gig Cheetah, 80megs/second
glibc 2.1.2
egcs
etc..

Server #2:
AMD Athlon 650
Asus K7M motherboard
512 Megs of 3.3 volt ECC SDRAM
3COM 3c905 ethernet card
Slackware 7
Kernel 2.2.14
Maxtor ATLAS 10k ULTRA WIDE SCSI 40megs/second
glibc 2.1.2
egcs
etc.

As you can see, these are basically identical setup except for the
CPU. First thing to point out, is that both mainboards are using the
same BIOS version.

Server #1 detects 512 megs of ram on bootup (BIOS), but
linux can only see 64 megs of ram unless I specify mem=512M
in the lilo configuration.

Server #2 detects 512 megs of ram on bootup, and linux
detects all the ram without extra parameters.

Server #1 is incredibly slow. Although it's got a faster processor and
a faster hard disk, it performs horribly compared to Server #2. Here's
an example of timing compilation of kernel 2.2.14 using:
time make bzImage
(after both kernels are configured identically and both servers are
idle with a load of 0.01)
Server #1
real    18m10.465s
user    16m52.590s
sys     1m3.790s
# free
          total       used       free     shared    buffers     cached
Mem   517540     306128     211412      16788     138436     145476
-/+ buffers/cache:      22216     495324
Swap:       265032          0     265032
# cat /proc/cpuinfo
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 2
model name      : AMD Athlon(tm) Processor
stepping        : 1
cpu MHz         : 755.315128
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
sep_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr 6 mce cx8 sep mtrr pge 14
cmov fcmov 17 22 mmx 24 30 3dnow
bogomips        : 752.03

Server #2
real    2m42.281s
user    2m13.650s
sys     0m13.760s
# free
             total       used       free     shared    buffers
cached
Mem:        517488     426344      91144      51512     140992
220848
-/+ buffers/cache:      64504     452984
Swap:       104384          0     104384
# cat /proc/cpuinfo
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 1
model name      : AMD-K7(tm) Processor
stepping        : 2
cpu MHz         : 654.599184
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
sep_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr 6 mce cx8 sep mtrr pge 14
cmov fcmov 22 mmx 30 3dnow
bogomips        : 652.08

Yes folks, the 750 was 16 minutes slower in compiling the kernel.
What's going on here? I haven't been able to figure it out. My guess
is that it's some kind of memory problem, but I'm completely baffled.
Any suggestions on how to fix this would be greatly appreciated.

If you've encountered this problem before and know how to fix it,
please email me your reply to this post.

Mark Steele
Systems Adminitrator
Belmont Web Services

The Athlon 650/700 use a cache speed that is 1/2 the core speed, but
750/800+ use 1/3 core speed.  Does this count for your differences?

As well, check BIOS and board rev levels - there may be something there...

This is odd.  Sounds like the two boards don't have identical ROM
BIOSes.  Try downloading the latest BIOS from ASUS and flash-update
Server #1.  Maybe #2 as well.

Quote:> Server #1 is incredibly slow. Although it's got a faster processor and
> a faster hard disk, it performs horribly compared to Server #2.

What about your SCSI cable lengths?  an LVD drive has a 12m cable length
limit, but only if the controller card is LVD.  Is it?  If it's using an
Ultra-Wide controller, then your SCSI cables should be 1.5m or less.  If
it's using a Fast-Wide controller, the cables should be 3m or less.

Quote:> Here's an example of timing compilation of kernel 2.2.14 using: time
> make bzImage
> (after both kernels are configured identically and both servers are
> idle with a load of 0.01)
> Server #1
> real    18m10.465s
> user    16m52.590s
> sys     1m3.790s
>  ....
> Server #2
> real    2m42.281s
> user    2m13.650s
> sys     0m13.760s

Hard drive problems, from bad termination or too-long cables could cause
extreme differences like these.  Do you find SCSI errors in any of your
log files?

Quote:> Server #1
> # free
>        total       used       free     shared    buffers     cached
> Mem   517540     306128     211412      16788     138436     145476
> -/+ buffers/cache:      22216     495324
> Swap:       265032          0     265032
>   ....
> Server #2
> # free
>              total       used       free     shared    buffers  cached
> Mem:        517488     426344      91144      51512     140992  220848
> -/+ buffers/cache:      64504     452984
> Swap:       104384          0     104384

The two systems are definitely using memory differently.  Although both
report the same amount of total RAM, the faster one is using more RAM
and less swap.  That could definitely hurt performance.

It apears that they are not as identicaly as they seem on the surface.

Quote:> Yes folks, the 750 was 16 minutes slower in compiling the kernel.
> What's going on here? I haven't been able to figure it out. My guess
> is that it's some kind of memory problem, but I'm completely baffled.
> Any suggestions on how to fix this would be greatly appreciated.

Check your SCSI setup.  A problem on the bus containing your swap
partition can severely impact everything.

Make sure RAM caching is properly turned on for Server #1.

Check the system's tuning.  It appears that System 1 is using less
memory and more swap for an identical task.  This bothers me.  The fact
that LILO only auto-detects 64M means that the installer only detected
64M.  It may have preset some system tuning values differently based on
that.

-- David

Incorrecto, 750/800 use 2/5 cache and no that can not explain a 16
minute difference. That kind of difference is probably due to defective
hardware. Fortunately with two systems you should be able to find the
problem by swapping components.

Sent via Deja.com http://www.deja.com/
Before you buy.