At 4:41 PM +0800 4/9/10, Min Miles Xu wrote:
Maurice Volaski wrote:
At 5:02 PM +0800 4/8/10, Min Miles Xu wrote:
Hi Maurice,
Did you notice how much time it took for the ping request packet
to arrive at the Windows machine and how much time for the echo
packet? And could you run the dtrace script in the attachment
while you ping? Just type "dtrace -s e1000g_tx.d"
At first, running ping from the OpenSolaris VM doesn't display
anything. Eventually, it does like so
64 bytes from thevaultinternal (172.16.1.2): icmp_seq=7944.
time=1549727.719 ms
64 bytes from thevaultinternal (172.16.1.2): icmp_seq=7947.
time=1547915.398 ms
64 bytes from thevaultinternal (172.16.1.2): icmp_seq=7951.
time=1546165.354 ms
64 bytes from thevaultinternal (172.16.1.2): icmp_seq=7952.
time=1546227.851 ms
64 bytes from thevaultinternal (172.16.1.2): icmp_seq=7954.
time=1545352.856 ms
64 bytes from thevaultinternal (172.16.1.2): icmp_seq=7955.
time=1545415.344 ms
On the Windows side, I see the pings being received and replied to.
This suggests to me it's the same problem described in the other
thread. And if you are right about that, it implies that even if I
could try a different NIC, it might still exhibit this behavior.
What's puzzling is why I can't seem to trigger it by running
multiple rsh threads and why many other users aren't experiencing
it. It's not a new problem. I've seen in way earlier builds,
possibly b110!
Hmm, according to the dtrace output, it takes no more than 1799 ms
for a kernel's ICMP round-trip. But the ping output showed
approaching 100 times longer. It seems to be something wrong inside
the VM. Does somebody else have any ideas?
I should have clarified that the actual ping time is as you say less
than 2 seconds or so, nothing like the crazy million ms being
misreported by OpenSolaris. Presumably, that is part of the bug. The
question is what is causing that and the already slow 1799 ms
response times.
Anyway, I have some interesting news to report. I tested with
OpenSolaris CIFS, the included Samba, and Samba 3.5.1 and they all
poison and cripple the interface. I also tried on the virtual switch
associated with a real NIC and it happens there too.
But since I couldn't get rsh from my Mac to do it, I wondered whether
the problem is somehow being triggered by some pattern intrinsic to
CIFS/SMB network traffic, so I am trying again by exporting the
OpenSolaris shares via NFS to a Linux VM on the same computer that in
turn re-exports them via Samba to the Windows VM. Guess what? This is
working!
--
Maurice Volaski, [email protected]
Computing Support, Rose F. Kennedy Center
Albert Einstein College of Medicine of Yeshiva University
_______________________________________________
networking-discuss mailing list
[email protected]
