VLAN definition loses packets on eth0

I have two identical Intel Core 2 Duo boxes. One is Fedora Core 6, uname -r shows 2.6.18-1.2798.fc6xen. The other box has Fedora Core 8 and uname -r says 2.6.23.1-42.fc8 (I wonder why there is no 'xen' ending? I downloaded x86_64 ISO DVD image, the same about FC6)
Both has identical configuration. The FC 6 is perfect, no issues. At FC 8 adding to /etc/rc.local
modprobe 8021q
vconfig add eth0 3
makes the box to lose 30-50% pings at eth0 (the other machine pings FC8).
Interestingly pinging eth0.3 address gives 0% lost frames.
What happened to FC8 it stopped working?

bolobeoAsked:
Who is Participating?
 
NopiusConnect With a Mentor Commented:
If nothing has changed except the kernel and everything was OK before, it _is_ a bug. ICMP lost packets may occur only because of problems in a hardware, in kernel driver or as a result of misconfiguration, but you didn't configure kernel parameters .

Since 2.6.21 where many changes in kernel API that have result to driver code, that may be rewritten.

That's a list of API changes: http://lwn.net/Articles/2.6-kernel-api/
 
If you say what driver are you using for eth0, we might figure out what changes had been made in this driver code. Even when no changes in a driver code, there might be changes in I/O scheduler or different default I/O scheduler in Fedora 8.

In any case if it was OK on Fedora 6 and bad on Fedora 8, I would never move production server to Fedora 8. Newer doesn't mean better.

0
 
NopiusCommented:
> makes the box to lose 30-50% pings at eth0 (the other machine pings FC8)

Where the other machine is located? In the  same LAN or not?
0
 
bolobeoAuthor Commented:
All the machines are in the same LAN, nothing in between.
0
Never miss a deadline with monday.com

The revolutionary project management tool is here!   Plan visually with a single glance and make sure your projects get done.

 
ravenplCommented:
I was thinking, that maybe it's the switch is misconfigured. maybe default vlan(guest) where eth0 is attached have some limits?
0
 
bolobeoAuthor Commented:
I tried 3 or 4 switches, two brands. No difference. Switching cables between switch ports of FC6 and FC8 changes nothing neither.
0
 
NopiusCommented:
What values do you have in /proc/sys/net/ipv4/icmp_ratelimit and /proc/sys/net/ipv4/icmp_ratemask on both systems?
0
 
bolobeoAuthor Commented:
/proc/sys/net/ipv4/icmp_ratemask is 6168 on both
/proc/sys/net/ipv4/icmp_ratelimit  is 1000 on FC8 and 250 on FC 6

0
 
NopiusCommented:
Did you tried to boot that machine, that have problems from some Linux LiveCD with the older kernel (like 2.6.18, just for test), configure the same vlans and try ping again?
0
 
bolobeoAuthor Commented:
No, but it run normal FC6 instalation before without problem.
I will download FC8 image once again and check SHA1 sum. We didn't do that before.
No better idea.
0
 
NopiusCommented:
So the problem  is in Linux kernel. It's far from the ideal and is constantly changing.
I'd recommend you to revert it back to FC6. The problem is most probably a bug.
0
 
arrkerr1024Connect With a Mentor Commented:
By the way, since no one else mentioned it - the xen ending on your fc6 kernel is support for xen virtualization.  If you don't run virtual machines through xen, you don't need that kernel.

They're both physical boxes right?
0
 
bolobeoAuthor Commented:
The boxes are physical indeed.
I just downloaded the FC6 image and I did not recall there was virtualization mentioned.
What I did now:
To be more like running FC6 I installed kernel-xen on FC8 (yum install kernel-xen) and booted from that.
Uname -r says 2.6.21-2952.fc8xen now.
But it is still the same.
I'm starting to believe it's a bug.
0
 
bolobeoAuthor Commented:
Today we tried another install from the scratch checking SHA1 sum before.
Nothing changed.
So we are coming back to FC6. We will check FC9alpha, just for curiosity.

Thanks Nopius and arrkerr1024

0
All Courses

From novice to tech pro — start learning today.