GNIC II on Linux 2.2.x

Jason Holmes jholmes@psu.edu
Mon Jul 26 10:26:40 1999


Greetings.

Is anyone successfully using the GNIC II (hamachi) cards on Linux 2.2.x
SMP?  We're seeing some strange problems.  For example, if the cards are
brought up with the default MTU of 1500, packets larger than the MTU
don't seem to fragment correctly.  I.e.,:

lion032g:~# ifconfig eth1
eth1      Link encap:Ethernet  HWaddr 00:E0:B1:04:20:35  
          inet addr:10.0.1.33  Bcast:10.0.1.255  Mask:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:3844481 errors:119 dropped:0 overruns:0 frame:119
          TX packets:3835079 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:100 
          Interrupt:19 

lion032g:~# ping -c 2 -s 3000 lion031g
PING lion031g.hpc.cac.psu.edu (10.0.1.32): 3000 data bytes

--- lion031g.hpc.cac.psu.edu ping statistics ---
2 packets transmitted, 0 packets received, 100% packet loss

If I drop the MTU to 1499, however, ping will work [other things still
die.]

lion032g:~# ifconfig eth1
eth1      Link encap:Ethernet  HWaddr 00:E0:B1:04:20:35  
          inet addr:10.0.1.33  Bcast:10.0.1.255  Mask:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1499  Metric:1
          RX packets:3844617 errors:119 dropped:0 overruns:0 frame:119
          TX packets:3835155 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:100 
          Interrupt:19 

lion032g:~# ping -c 2 -s 3000 lion031g
PING lion031g.hpc.cac.psu.edu (10.0.1.32): 3000 data bytes
3008 bytes from 10.0.1.32: icmp_seq=0 ttl=255 time=1.3 ms
3008 bytes from 10.0.1.32: icmp_seq=1 ttl=255 time=0.9 ms

--- lion031g.hpc.cac.psu.edu ping statistics ---
2 packets transmitted, 2 packets received, 0% packet loss
round-trip min/avg/max = 0.9/1.1/1.3 ms

If I drop the MTU as far as 1400, I can run some of the standard MPI
tests [PingPing, PingPong] successfully, but the performance isn't what
it should be.

This was all with PacketEngines Powerrail 5300 Switch.  The same thing
happens if we swap in an FDR.  The driver being used is the version 0.14
found at http://www.nscl.msu.edu/~kasten/perf/hamachi/src/, not compiled
for SMP [if I add the flags '-D__SMP__ -DCONFIG_X86_LOCAL_APIC', it
compiles ok, but things died a lot and I don't have the error messages
on me at the moment.]  Everything has been compiled with gcc 2.7.2.3.

So, has anyone else run across problems such as the one above?  If not,
can you tell me what kernel and what version of the driver are you
using?  If so, have you found a solution?

Thanks,

--
Jason Holmes
 | To unsubscribe, send mail to Majordomo@cesdis.gsfc.nasa.gov, and within the
 |  body of the mail, include only the text:
 |   unsubscribe this-list-name youraddress@wherever.org
 | You will be unsubscribed as speedily as possible.