FAQ Search Today's Posts Mark Forums Read
» Video Reviews

» Linux Archive

Linux-archive is a website aiming to archive linux email lists and to make them easily accessible for linux users/developers.


» Sponsor

» Partners

» Sponsor

Go Back   Linux Archive > CentOS > CentOS

 
 
LinkBack Thread Tools
 
Old 07-14-2010, 03:17 PM
Whit Blauvelt
 
Default Linux Kernel Physical Interface Limit

On Wed, Jul 14, 2010 at 09:51:51AM -0500, Tim Nelson wrote:

> Even if the limit were lower, such as 10 physical interfaces as mentioned
> before, I have to imagine that the host system would have issues dealing
> with the number of interrupts needed to *PROPERLY* service all of those
> interfaces in addition to the other system hardware.

There may (or may not) be another problem. As of a couple of years ago, on
some Linux variants (didn't try RHEL/CentOS), I was having trouble even
getting 6 NICs (on 3 cards) to work if I had IPv6 turned on. 4 NICs worked
fine.

Filed some bug reports, and it was evident from the response that very, very
few Linux users ever go > 4 eth's on a system. Thus the lack of properly
debugged IPv6 support for that then. Fortunately I don't (yet) need IPv6.
When I do, it'll be curious to see if the bug is still there.

Whit
_______________________________________________
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos
 
Old 07-14-2010, 03:52 PM
Benjamin Franz
 
Default Linux Kernel Physical Interface Limit

On 07/14/2010 08:17 AM, Whit Blauvelt wrote:
> On Wed, Jul 14, 2010 at 09:51:51AM -0500, Tim Nelson wrote:
>
>
>> Even if the limit were lower, such as 10 physical interfaces as mentioned
>> before, I have to imagine that the host system would have issues dealing
>> with the number of interrupts needed to *PROPERLY* service all of those
>> interfaces in addition to the other system hardware.
>>
> There may (or may not) be another problem. As of a couple of years ago, on
> some Linux variants (didn't try RHEL/CentOS), I was having trouble even
> getting 6 NICs (on 3 cards) to work if I had IPv6 turned on. 4 NICs worked
> fine.
>
> Filed some bug reports, and it was evident from the response that very, very
> few Linux users ever go> 4 eth's on a system. Thus the lack of properly
> debugged IPv6 support for that then. Fortunately I don't (yet) need IPv6.
> When I do, it'll be curious to see if the bug is still there.
>

I've got six machines with 6 Gb interfaces (two on motherboard, 4 on a
card) right now (the design called for 3 bonded pairs on separate nets
for redundancy). I haven't tried IPV6 on them. I had 'issues' with
bonding and VMs though.

--
Benjamin Franz
_______________________________________________
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos
 
Old 07-14-2010, 03:57 PM
Whit Blauvelt
 
Default Linux Kernel Physical Interface Limit

On Wed, Jul 14, 2010 at 08:52:02AM -0700, Benjamin Franz wrote:

> > Filed some bug reports, and it was evident from the response that very, very
> > few Linux users ever go> 4 eth's on a system. Thus the lack of properly
> > debugged IPv6 support for that then. Fortunately I don't (yet) need IPv6.
> > When I do, it'll be curious to see if the bug is still there.
>
> I've got six machines with 6 Gb interfaces (two on motherboard, 4 on a
> card) right now (the design called for 3 bonded pairs on separate nets
> for redundancy). I haven't tried IPV6 on them. I had 'issues' with
> bonding and VMs though.

My trouble didn't involve bonding or VMs, just using 5 interfaces at once
(dual-homed WAN, DMZ, LAN, direct crossover failover backup). Having IPv6
turned on was screwing up using more than 4 for IPv4. Suspect it was a
kernel limitation that in likelihood is since fixed.

Whit
_______________________________________________
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos
 
Old 07-14-2010, 05:30 PM
JohnS
 
Default Linux Kernel Physical Interface Limit

On Wed, 2010-07-14 at 08:52 -0700, Benjamin Franz wrote:

> I've got six machines with 6 Gb interfaces (two on motherboard, 4 on a
> card) right now (the design called for 3 bonded pairs on separate nets
> for redundancy). I haven't tried IPV6 on them. I had 'issues' with
> bonding and VMs though.
---
Can you give me "sar -I SUM" the last timed entry intr/s? How many
CPUs?
I'm not questioning you but on the curious side.

John

_______________________________________________
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos
 
Old 07-14-2010, 06:31 PM
Benjamin Franz
 
Default Linux Kernel Physical Interface Limit

On 07/14/2010 10:30 AM, JohnS wrote:
> On Wed, 2010-07-14 at 08:52 -0700, Benjamin Franz wrote:
>
>
>> I've got six machines with 6 Gb interfaces (two on motherboard, 4 on a
>> card) right now (the design called for 3 bonded pairs on separate nets
>> for redundancy). I haven't tried IPV6 on them. I had 'issues' with
>> bonding and VMs though.
>>
> ---
> Can you give me "sar -I SUM" the last timed entry intr/s? How many
> CPUs?
> I'm not questioning you but on the curious side.
>

On the heaviest loaded machine:

10:10:01 AM sum 1637.48
10:20:01 AM sum 1640.73
10:30:01 AM sum 1653.58
10:40:01 AM sum 1617.78
10:50:01 AM sum 1727.97
11:00:01 AM sum 1767.88
11:10:01 AM sum 1798.93
11:20:01 AM sum 1782.14

Average: INTR intr/s
Average: sum 1365.55

This is on a dual processor machine with a total of 8 cores.

The highest I see on any of the machines for the last 24 hours is a
brief (one ten minute interval) peak of 5300 intr/second during system
backups and nothing over 3000 otherwise.

--
Benjamin Franz

_______________________________________________
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos
 
Old 07-14-2010, 08:43 PM
Gordon Messmer
 
Default Linux Kernel Physical Interface Limit

On 07/13/2010 05:51 PM, William Warren wrote:
> I think it's baloney mainly because i can't find a mention of it
> anywhere. Is there REALLY a limit on the number of physical network
> interfaces in the Kernel?

Most of the limits on the number of any device come from the major/minor
numbers on the device nodes. Since ethernet adapters don't have device
nodes, I'd be surprised if there were any practical limit.

I'm able to create at least 1025 VLANs on my ethernet adapter, and VLANs
are "real" interfaces in the kernel.
_______________________________________________
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos
 
Old 07-15-2010, 08:14 AM
Emmanuel Noobadmin
 
Default Linux Kernel Physical Interface Limit

On 7/14/10, William Warren <hescominsoon@emmanuelcomputerconsulting.com> wrote:
> On 7/14/2010 1:16 AM, Emmanuel Noobadmin wrote:
>> Googling up a really old 2005 newsgroup thread says some people had 24
>> physical NIC (6x 4quad) in a system before and one person vaguely
>> remembers a hard 256 limit which would make sense if physical
>> interface count is a byte value.
> can you give me a link to that thread? My googling skills are
> apparently not up to snuff to find that..

2001 poster says kernel 2.4 = 16 at least
http://groups.google.com/group/comp.os.linux.hardware/browse_thread/thread/fbb020bfcc0abd/f25e760b552c9246?pli=1

2005 poster says he's seen 6x quad in a Linux server
http://linux.derkeiler.com/Newsgroups/comp.os.linux.hardware/2005-01/0431.html
_______________________________________________
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos
 

Thread Tools




All times are GMT. The time now is 08:04 AM.

VBulletin, Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO ©2007, Crawlability, Inc.
Copyright 2007 - 2008, www.linux-archive.org