PATCH: Advanced Ping Template: accurate ping results

gashalot · Joined: 07 Sep 2005 Posts: 9

After migrating from a Windows cacti host to a Linux host, I noticed that the internal ICMP ping functionality in the <cacti>/lib/ping.php library does not accurately report network latency.

In the case of my network, Cacti's homegrown ICMP ping was reporting response times of 2-3 ms for host located approximately 2500 miles away. Using the standard (dist*2)/100 latency calculation, the absolute minimum latency on this link should be approximately 60ms. I tested the same host using the native ping, and results were reported correctly.

I decided against patching <cacti>/lib/ping.php, as it includes UDP and SNMP ping functionality used to detect dead hosts during the pollinc cycle, and checking host "aliveness" does not require accurate latency data.

Instead I have modified <cacti>/scripts/ss_fping.php script to report "true" network performance and standard deviation by using the PEAR Net_Ping and Math_Stats modules.

These modules have been tested on Debian/sarge running php5, and have been validated via the script server. Note that you may need to manually install the Net_Ping and Math_Stats module. Run "pear list" on your system to determine if this is necessary.

Installation Instructions
Replace <cacti>/scripts/ss_fping.php with the file included in the library, or apply the diff.

NOTE: the Net_Ping v2.4 has a critical flaw that causes ping output on certain Linux systems to be parsed incorrectly. I have included a patched version of the Ping.php (located in the <pear>/Net folder) as well.[/b]

TheWitness · Posted: Sat Sep 24, 2005 8:30 pm Post subject:

The problem with ss_fping.php (as I see it, never tested, don't run *nix) with a Unix poller is that the php binary must have asroot privileges to create a raw socket. TCP pings to something like port 80 should work.

Otherwise, it returns bogus data after likely an error. It works AOK in Windows. It also may be issues with the timeout in the socket_set_option() call that does not appear to work 100% in various versions of PHP.

I have updated, ss_fping for the correct standard deviation and inclulded some timeout detection as there are issues when your script timeout is low and you are getting quite a bit of packet loss. I have not posted it though.

TheWitness

gashalot · Joined: 07 Sep 2005 Posts: 9

It may be a UNIX-related issue. I recently migrated from a Win2k platform, where the ss_fping worked correctly (or at least obeyed the laws of physics). Only when I switched to UNIX did I start to see suspect latency reported.

For the record, the pollers do run as root on our UNIX boxes, but they behaved strangely when using a UDP ping as well... I'm not sure what would be causing the problem.

In the future it may be worth investigating replacing the cacti homebrew ICMP with Net_Ping (after I get some of the big bugs with it patched). Not only would it eliminate a potential cause of headache (such as the physically impossible ping responses), but it would allow non-root pollers on UNIX to use ICMP when their ping executable was installed setuid (which most UNIX installations do).

Food for thought... For now the patches are working "correctly" for us once again!

sizulku · Cacti User Joined: 04 Nov 2002 Posts: 102 Location: ACEH

After applying patch, I'm having problem with the script output as shown below;

[root@ws159132 scripts]# php ss_fping.php www.yahoo.com 10 ICMP
sh: /usr/bin/cpp: No such file or directory
PHP Warning: preg_match() expects parameter 2 to be string, array given in /usr/share/pear/Net/Ping.php on line 685
Content-type: text/html
X-Powered-By: PHP/4.3.9

 
Warning: preg_match() expects parameter 2 to be string, array given in /usr/share/pear/Net/Ping.php on line 685 
min:678.1700 avg:691.8970 max:708.3660 dev:10.5956 loss:0.0000
[root@ws159132 scripts]#

The ping result now is correct however I need to remove the warning output of the script. FYI, I'm using:

CentOS release 4.3 (Final)
Cacti 0.8.6h
cactid 0.8.6f-1
Net_ping 2.4
Math_Stats

Would you please let me know how? any help would be great.

TheWitness · Posted: Fri Aug 18, 2006 2:19 pm Post subject:

Hello!

The following code is not ours although it may look similar. Don't know, never looked at it:

sizulku · Cacti User Joined: 04 Nov 2002 Posts: 102 Location: ACEH

would you please take a look at the code? I've tried many ping's script for linux but only this patch able to produce the correct output.

TheWitness · Posted: Sat Aug 19, 2006 11:03 am Post subject:

Unfortunately, you are the only one to be able to resolve this bug as I do not have your system. The error is pretty clear, you just have to find out what's different and then open a bug report with the author to fix.

TheWitness

TheWitness · Posted: Sat Aug 19, 2006 11:05 am Post subject:

If you are using the latest version of Cacti's builtin ping.php, you should be able to use ICMP ping BTB. In other words you should not require this hack.

TheWitness

sizulku · Cacti User Joined: 04 Nov 2002 Posts: 102 Location: ACEH

Hi TheWitness! Thanks for your reply, I greatly appreciate it.

Now I've re-install my system using cacti-0.8.6h from download url http://www.cacti.net/downloads/cacti-0.8.6h.tar.gz.

The output of Cacti's builtin ping.php still incorrect. Let's see;

[root@localhost scripts]# ping -c 15 www.yahoo.com
PING www.yahoo.akadns.net (209.131.36.158) 56(84) bytes of data.
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=0 ttl=57 time=684 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=1 ttl=57 time=713 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=2 ttl=57 time=715 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=3 ttl=57 time=690 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=4 ttl=57 time=725 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=5 ttl=57 time=845 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=6 ttl=57 time=712 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=7 ttl=57 time=672 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=8 ttl=57 time=711 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=9 ttl=57 time=682 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=10 ttl=57 time=685 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=11 ttl=57 time=707 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=12 ttl=57 time=714 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=13 ttl=57 time=684 ms
64 bytes from f1.www.vip.sp1.yahoo.com (209.131.36.158): icmp_seq=14 ttl=57 time=709 ms

--- www.yahoo.akadns.net ping statistics ---
15 packets transmitted, 15 received, 0% packet loss, time 14249ms
rtt min/avg/max/mdev = 672.605/710.354/845.120/39.172 ms, pipe 2

[root@localhost scripts]# php -q ss_fping.php www.yahoo.com 15 ICMP
min:28.3600 avg:188.9779 max:288.9500 dev:80.4398 loss:6.6667

TheWitness · Posted: Sat Aug 19, 2006 2:01 pm Post subject:

What is your host's timeout in Cacti set to?

TheWitness

sizulku · Cacti User Joined: 04 Nov 2002 Posts: 102 Location: ACEH

Host's timeout in cacti was set using default value 400 and I've fix the script to produce accurate ping result.

From the script server command line I get a result:

TheWitness · Posted: Sun Aug 20, 2006 12:54 am Post subject:

Not too sure yet, but I have updated lib/ping.php in SVN BRANCH_0_8_6 to reflect more accurate results if the Cacti timeout is set too low.

Here it is below as well.

TheWitness

sizulku · Cacti User Joined: 04 Nov 2002 Posts: 102 Location: ACEH

Thanks for the update. Just been replaced ping.php with the new one and seemed I got an accurate result now.

Result from script server command line:

TheWitness · Posted: Sun Aug 20, 2006 6:51 am Post subject:

I am embarrased to say that the RRDfile does not store results greater than 500ms. So, edit the PING Data Template and change the max value for AVG, MAX, and MIN to 10000, delete your RRDfile and you should be Ok.

TheWitness

sizulku · Cacti User Joined: 04 Nov 2002 Posts: 102 Location: ACEH

Thanks, it work now. I'm still wondering on how packet loss was calculated since on the graph I found many pkt loss but it did not appear when I'm running script from command line or poller log.