2.0.35-0.2 from digital - weird net interface (DE4x5) latency

[abel bfr co il (Alexander L. Belikoff)]

Hello everybody,

I'm seeing weird network interface latency, which happens specifically
when I use 2.0.35-0.2 from ftp.digital.com. I am trying to mount
certain NFS partitions (from two other Linux Alphas) at the end of
/etc/rc.d/rc.local. This works fine on my ancient 2.0.30 kernel with
millions of ancient patches. Yet it doesn't work with the new kernel:

	mount clntudp_create: RPC: Port mapper failure - RPC: Unable to send


It is not specific to mount/RPC. I tried to use rsync from rc.local
and got connection problems as well.

I tried a test - to send a couple (literally) PING packets to each
host involved before doing the mounts - same problem:


PING foo.bar.com (...): 56 data bytes

--- foo.bar.com ping statistics ---
5 packets transmitted, 0 packets received, 100% packet loss


PING foo2.bar.com (...): 56 data bytes

--- foo2.bar.com ping statistics ---
5 packets transmitted, 0 packets received, 100% packet loss


However, by the time I log in, the network connection seems to be
fine. Moreover, I found a temporary cure for the problem (read "an
ugly hack") by sending 2 PINGs 10 times each to each of the hosts
involved before doing anything else. This seems to fix the problem...

I'll appreciate any help or ideas or suggestions.

Thanks in advance,


PS. It is important to note, that the old kernel used tulip.c as a net 
driver, while the new one is using de4x5.c. In both cases the drivers
are compiled in the kernel (not as modules).


PPS. This is the machine configuration:

=========================================================================

* LX164 motherboard, 128Mb of RAM, 128Mb of swap space

* RedHat 4.2

* SRM console

* headless machine (no kbd/mouse/videocard) - serial console


* SCSI controller: Intraserver NCR 53c8xx based. Detected as:

ncr53c8xx: at PCI bus 0, device 5, function 0
ncr53c8xx: PCI_LATENCY_TIMER=0, bursting should'nt be allowed.
ncr53c8xx: PCI_CACHE_LINE_SIZE not set, features based on CACHE LINE SIZE not u.
ncr53c8xx: 53c875 detected with Symbios NVRAM
ncr53c875-0: rev=0x04, base=0x9001000, io_port=0x9000, irq=18
ncr53c875-0: NCR clock is 40218KHz, 40037KHz
ncr53c875-0: Symbios format NVRAM, ID 7, Fast-20, Parity Checking
ncr53c875-0: initial SCNTL3/DMODE/DCNTL/CTEST3/4/5 = (hex) 00/00/00/00/00/00
ncr53c875-0: final   SCNTL3/DMODE/DCNTL/CTEST3/4/5 = (hex) 05/42/20/00/08/24
ncr53c875-0: on-chip RAM at 0x9002000
ncr53c875-0: requesting shared irq 18 (dev_id=0xfffffc0007fec0c0)
ncr53c875-0: resetting, command processing suspended for 2 seconds
ncr53c875-0: restart (scsi reset).
ncr53c875-0: enabling clock multiplier
ncr53c875-0: Downloading SCSI SCRIPTS.
scsi0 : ncr53c8xx - revision 3.0i
scsi : 1 host.
ncr53c875-0: command processing resumed
  Vendor: SEAGATE   Model: ST39102LW         Rev: 0004
  Type:   Direct-Access                      ANSI SCSI revision: 02
Detected scsi disk sda at scsi0, channel 0, id 0, lun 0
ncr53c875-0-<0,0>: tagged command queue depth set to 8
scsi : detected 1 SCSI disk total.
ncr53c875-0-<0,*>: WIDE SCSI (16 bit) enabled.
ncr53c875-0-<0,*>: FAST-20 WIDE SCSI 40.0 MB/s (50 ns, offset 15)
SCSI device sda: hdwr sector= 512 bytes. Sectors= 17783240 [8683 MB] [8.7 GB]

(we force fast SCSI by doing 'echo "setsync 0 12" > /proc/scsi/ncr53c8xx/0')


* DEC DS21140 Tulip network card running in 100Mbps full duplex:

eth0: DC21140 at 0x8800 (PCI bus 0, device 9), h/w address 00:c0:f0:31:f1:72,
eth0: Using generic MII device control. If the board doesn't operate, 
please mail the following dump to the author:

MII device address: 1
MII CR:  3000
MII SR:  7809
MII ID0: 2a8
MII ID1: 151
MII ANA: 1e1
MII ANC: 0
MII 16:  0
MII 17:  0
MII 18:  1

      and requires IRQ19 (provided by PCI BIOS).
de4x5.c:V0.5351 1998/7/5 davies@maniac.ultranet.com


=========================================================================

-- 
Alexander L. Belikoff
Bloomberg L.P. / BFM Financial Research Ltd.
abel@vallinor4.com, abel@bfr.co.il