On Mon, 02 Jul 2012 07:57:47 +
Bart Van Assche bvanass...@acm.org wrote:
On 07/01/12 11:02, Vincent Bernat wrote:
NetSNMP contains several use of a netlink socket to gather information
from the kernel for Linux:
agent/mibgroup/etherlike-mib/data_access/dot3stats_linux.c
On Wed, 16 Mar 2011 11:13:25 -0400
Robert Story rst...@freesnmp.com wrote:
On Fri, 11 Mar 2011 09:10:01 -0800 Stephen wrote:
SH The netlink one is a bit problematic, in that it unconditionally
replaces the
SH old code, and we need to support older systems which might not have
netlink.
On Fri, 11 Mar 2011 08:46:51 -0500
Robert Story rst...@freesnmp.com wrote:
On Mon, 28 Feb 2011 12:37:17 -0800 Stephen wrote:
SH These patches fix the problems with walking the forwarding mib when a
SH router has full BGP feed.
Excellent! I'm glad someone finally had the time to fix this.
On Fri, 11 Mar 2011 08:46:51 -0500
Robert Story rst...@freesnmp.com wrote:
On Mon, 28 Feb 2011 12:37:17 -0800 Stephen wrote:
SH These patches fix the problems with walking the forwarding mib when a
SH router has full BGP feed.
Excellent! I'm glad someone finally had the time to fix this.
On Fri, 11 Mar 2011 18:46:56 +0100
Bart Van Assche bvanass...@acm.org wrote:
On Fri, Mar 11, 2011 at 6:42 PM, Stephen Hemminger
shemmin...@vyatta.comwrote:
On Fri, 11 Mar 2011 08:46:51 -0500
Robert Story rst...@freesnmp.com wrote:
On Mon, 28 Feb 2011 12:37:17 -0800 Stephen wrote
The route table maybe large (600K) on a fully connected router,
and the table_container has terrible performance (N^2 on inserts).
Since this container is only a temporary container used while filtering
routes from the kernel, switch to a FIFO since the next pass will
reorder routes anyway.
---
These patches fix the problems with walking the forwarding mib when a
router has full BGP feed.
See
http://linux-network-plumber.blogspot.com/2011/02/net-snmp-ip-forward-table-performance.html
for the full saga.
--
---
agent/mibgroup/ip-forward-mib/inetCidrRouteTable/inetCidrRouteTable_interface.c
|2 +-
agent/mibgroup/ip-forward-mib/ipCidrRouteTable/ipCidrRouteTable_interface.c
|2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
---
Use netlink (instead of /proc) to read route table on linux.
This is more efficient and has the advantage of returning interface index
not interface name which avoids lots of unnecessary lookups.
Both IPv4 and IPv6 share the same code as well.
This code also handles multipath and ipv6 routes
Setting the ALLOW_DUPLICATES fixes the insert performance problem.
and causes only one sort to occur on the first access.
---
agent/mibgroup/ip-forward-mib/inetCidrRouteTable/inetCidrRouteTable_interface.c
|6 +-
On Tue, 30 Nov 2010 07:54:32 -0800
Andrus, Brian Contractor bdand...@nps.edu wrote:
Not sure how to check the cache interval.
Made the changes. Not seeing anything special to help figure it out. Only new
entry in log was at restart of the service:
snmpd[14775]: registered debug token
It looks like it can take 5 to 6 minutes for SNMP daemon to
decide interface has really gone away. It will take 30 seconds
for it to decide to rescan the cache (on linux), then it marks it down
for 5 minutes.
This seems bogus and excessive, what really should happen:
* snmpd should have
On Thu, 09 Sep 2010 10:19:27 -0700
Wes Hardaker harda...@users.sourceforge.net wrote:
On Tue, 7 Sep 2010 15:17:01 -0700, Stephen Hemminger
shemmin...@vyatta.com said:
SH It is a zero-length string unless admin has set it through
SH ip link set dev eth0 alias 'my network interface'
SH
On Fri, 3 Sep 2010 14:27:02 -0400
Robert Story rst...@freesnmp.com wrote:
On Fri, 3 Sep 2010 10:35:05 -0700 Stephen wrote:
SH This is the correct patch.
SH
SH Subject: if-mib: add support for ifAlias
SH
SH Linux has a way of recording an ifAlias value for each interface.
SH The value is
On Tue, 07 Sep 2010 15:08:26 -0700
Wes Hardaker harda...@users.sourceforge.net wrote:
On Tue, 7 Sep 2010 11:22:08 -0700, Stephen Hemminger
shemmin...@vyatta.com said:
SH I have fix to make it keep overrides. But I did not implement
SH setting kernel ifalias because on most systems snmpd
This is the correct patch.
Subject: if-mib: add support for ifAlias
Linux has a way of recording an ifAlias value for each interface.
The value is configurable with iproute utilities or sysfs.
---
agent/mibgroup/if-mib/data_access/interface.c |6
On Fri, 21 May 2010 19:04:43 -0400
Doug Manley doug.man...@gmail.com wrote:
Hi, team!
I've submitted numerous patches to the net-snmp project for errors so
esoteric that it took days to find and understand them.
In my painful searches through the commentless code, I built up an
On Thu, 4 Mar 2010 10:05:57 -0500
Robert Story rst...@freesnmp.com wrote:
On Wed, 3 Mar 2010 22:30:40 -0800 Stephen wrote:
SH I see the patch. It allows clientaddr to be in snmpd.conf without the
[snmp]
SH prefix. But clientaddr is a snmp token, and does work in snmpd.conf when
SH
On Thu, 4 Mar 2010 00:50:04 -0500
Robert Story rst...@freesnmp.com wrote:
On Tue, 2 Mar 2010 09:13:29 -0800 Stephen wrote:
SH I suspect you're putting clientaddr in snmpd.conf instead of snmp.conf.
If you
SH want to do that, you have to prefix it for it to be recognized as a
snmp.conf
On Thu, 4 Mar 2010 00:50:04 -0500
Robert Story rst...@freesnmp.com wrote:
On Tue, 2 Mar 2010 09:13:29 -0800 Stephen wrote:
SH I suspect you're putting clientaddr in snmpd.conf instead of snmp.conf.
If you
SH want to do that, you have to prefix it for it to be recognized as a
snmp.conf
On Tue, 2 Mar 2010 10:54:53 -0500
Robert Story rst...@freesnmp.com wrote:
On Wed, 13 Jan 2010 19:37:45 -0800 Stephen wrote:
SH It is documented in snmp.conf(5) that setting clientaddr is supposed
SH to be used as trap source address, but it doesn't work. The issue is
SH that the configuration
On Thu, 11 Feb 2010 12:36:39 +0100
Bart Van Assche bvanass...@acm.org wrote:
On Thu, Feb 11, 2010 at 12:22 PM, Josef Moellers
josef.moell...@ts.fujitsu.com wrote:
Stephen Hemminger wrote:
On Wed, 10 Feb 2010 12:02:44 +0100
Josef Moellers josef.moell...@ts.fujitsu.com wrote
On Fri, 12 Feb 2010 11:52:04 -0800
Wes Hardaker harda...@users.sourceforge.net wrote:
On Tue, 26 Jan 2010 01:27:48 -0800, Stephen Hemminger
shemmin...@vyatta.com said:
Checking the code in 'container.c', it looks as if binary_array
is the default approach for table_container
SH
On Wed, 10 Feb 2010 12:02:44 +0100
Josef Moellers josef.moell...@ts.fujitsu.com wrote:
Dave Shield wrote:
On 10 February 2010 09:23, Josef Moellers josef.moell...@ts.fujitsu.com
wrote:
I'd like to amend the etherlike-mib code for Linux
Any comment/objection while I locally
On Tue, 26 Jan 2010 08:22:38 +
Dave Shield d.t.shi...@liverpool.ac.uk wrote:
2010/1/25 Stephen Hemminger shemmin...@vyatta.com:
Some of the SNMP tables can get large (like route table), and the existing
table storage implementation has linear search on insert. Has anyone
done
Some of the SNMP tables can get large (like route table), and the existing
table storage implementation has linear search on insert. Has anyone
done an alternative that uses something smarter (tree, hash, b-tree)?
This would fix a number of issues when using net-snmp with a router
with a full
The default linkUpDownNotification trap, polls for link events every 60 seconds.
This is both wasteful (link events are rare), and too slow (delaying 60 seconds
in notifying admin). Has anyone done a patch to use an event based mechanism.
Something that uses existing Linux netlink link state
On Mon, 21 Dec 2009 21:18:54 -0500
lanas la...@securenet.net wrote:
All,
Just to let you know that 5.4.1 as shipped by Debian applies a patch
# 1775124 that contains two strdup() inside a sprintf() statement in
the netsnmp_udp_fmtaddr() method (plus one more)
See:
The code parsing cpuinfo strings could read past end of buffer.
Avoid this problem by using only part of the allocated size
and null terminating.
Add check for read error.
---
agent/mibgroup/hardware/cpu/cpu_linux.c | 33 +-
1 files changed, 23 insertions(+), 10
--
--
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based
Valgrind can't figure out that speed will be set by SIOCETHTOOL,
so just initialize it.
---
.../mibgroup/if-mib/data_access/interface_linux.c |1 +
1 files changed, 1 insertions(+), 0 deletions(-)
--- a/agent/mibgroup/if-mib/data_access/interface_linux.c 2009-03-16
16:44:07.608956089
Same proble as other proc scanning, need to terminate string
before calling strstr.
---
agent/mibgroup/hardware/memory/memory_linux.c |7 ---
1 files changed, 4 insertions(+), 3 deletions(-)
--- a/agent/mibgroup/hardware/memory/memory_linux.c 2009-03-16
16:44:16.512954764 -0700
+++
The code for reading /proc files guesses at an initial size buffer,
then has to retry if the buffer isn't big enough. This slows boot
since it has to repeat several time. It was a penny wise (save memory)
but pound foolish (cpu) choice in initial design. Instead choose size
that gets rounded up to
The interface MIB is restricted at present to 10/100/1000 Mbit
ethernet interfaces on Linux. Ethtool now supports 10G and other speeds.
The convention is to check only for 0 and -1 as unknown speeds
diff --git a/agent/mibgroup/if-mib/data_access/interface_linux.c
On Sun, 5 Oct 2008 10:36:16 -0700 (PDT)
Amit Kumar [EMAIL PROTECTED] wrote:
Hi All,
how to restrict the manager to contact to particular agent Ip
address ?
is agentaddress in snmpd.conf is for that ?
I have a requirement , if the snmp agent IP 255.255.255.255 then any snmp
On Fri, 05 Sep 2008 21:48:10 +0200
Magnus Fromreide [EMAIL PROTECTED] wrote:
On Thu, 2008-09-04 at 14:18 -0700, Stephen Hemminger wrote:
Reduce snmpd memory usage.
The route table can grow very large and existing code is a major
performance problem since it wastes half of its memory
Rework how route table is read to use netlink interface.
This solved major performance problems when using SNMP on full gateway
table (256K routes).
The old code read /proc/net/route and then looked up the mapping from
name eth0 to ifindex for each entry. Much more efficient to use netlink
to
Reduce snmpd memory usage.
The route table can grow very large and existing code is a major
performance problem since it wastes half of its memory and copies
each time.
1. Grow expanding table by 1.5 rather 2x per step
Perhaps the increment should just be PAGESIZE
2. Use realloc rather
In testing Vyatta router, we found snmpd gets bogged down looking
up routing table. Converting the code to use netlink fixed that.
http://sourceforge.net/tracker/index.php?func=detailaid=1991406group_id=12694atid=312694
-
39 matches
Mail list logo