Hitting the ceiling... Post your stats here! :)

TPC · Post by **TPC** » Thu Oct 07, 2004 6:15 pm

Lets' share some stats.

On a P4 1.6mhz with 256mb ram, polling 730 objects (via snmp1).
the graph's starting to get choppy when someone views the graph at the same time it's polling. then again even with 100 to 150 objects.. the cpu utilisation has gone skyrocket to 100% everytime it polls.. how about for you guys?

P4 1.6mhz
256mb ram
730 Objects

OS win2k3
MySql 4.x
RRDTool 1.048
Cacti 0.8.6b
Perl 5.x
PHP 4.6.9? or something like that..
Running Cactid as poller (1 thread.. should i increase it? seeing as the cpu is hitting 100% anyhow)

Thinking of switching over everything to my linux machine to see if there's any significant performance boost... any thoughts on the matter?

JosephA · Post by **JosephA** » Thu Oct 07, 2004 7:26 pm

Funny, I was just about to post to ask just how many objects could be polled on one system. We want to poll about 4000-5000 objects, is anybody currently doing this?

We have this running on Fedora Core 2 in a VM with dual 2.8Ghz P4s and 512 ram.

MySQL 3.23.58
RRDtool 1.0.45
Cacti 0.8.6b
Perl 5.8.0
PHP 4.2.2

We are currently polling about 1000 objects on a slower system (don't have the exact specs right now) with no problems.

Post by **TheWitness** » Thu Oct 07, 2004 8:03 pm

TPC,

What type of hosts are your polling.

1) Windows NT HostMIB (CPU & Disk)?
2) Was this an upgrade?
3) Scripts? What kind?
4) How many concurrent processes? cmd.php does not recognize them.

If you can send me a copy of you database, I can answer more clearly. To export, "c:\mysql\bin\mysqldump -u=root -p cacti > mydump.sql".

Thanks,

TheWitnes

TPC · Post by **TPC** » Fri Oct 08, 2004 12:50 am

Heya TheWitness,

I wasn't really asking .. a question

just curious as to how much cacti can poll before starting to wither on different kind of systems

as in looking for the limits... heh.

I am polling basically network routers (ciscos and junipers), all within very low latency, from a fresh reinstallation, using the snmp interface stats script + one of my own to count dynamic mac addresses, and using 2 concurrent processes..

Thanks for replying anyhow TheWitness

Webmassa · Post by **Webmassa** » Fri Oct 08, 2004 1:55 am

At the moment we poll 500 objects from 8 different branch offices (VPN) and since cacti 0.8.6b we get no gaps in the graphs.

I use
- a dual PII Xeon (2x450Mhz)
- 3Com 3C905 Networkcard
- AMD PcNet32 Fast+ Networkcard --> I tried both, getting the same good results
- Adaptec AIC 7880 SCSI with Seagate ST34501W and ST39102LW connected
- 384 MB RAM
- OS: Sentinix Linux with self compiled Kernel 2.4.27
- Poll with cmd.php an 2 Processes
- Runtime 130s

So if I would scan twice the amount of objects (e.g. 1000) I think that cm.php is not suitable on our host (130s *2 = timeout *g*)
Maybe a Gigabit NIC is needed?

Regards...Webmassa

Post by **TheWitness** » Fri Oct 08, 2004 7:22 pm

Webmassa,

Check your CPU utilization. If it is less than 100%, then you can increase your number of concurrent processes. It is definately NOT the NIC.

Also, if you are polling Windows NT systems using the legacy HostMib queries for disk and cpu, you can also decrease CPU significantly and increase speed by converting to the Script Server. If this is the case, you should review the online documentation as to how to complete that migration.

In future releases, we will be supporting multiple pollers. Just be patient or upgrade to a faster processor, or start to use CACTID.

TheWitness

TPC · Post by **TPC** » Sat Oct 09, 2004 2:30 am

i just been adding tons more objects to my cacti.. now it has over >1200 objects

polling seems okay.. EXCEPT when i go make some adjustments and add in new objects and ask the new object to run some snmp query, i suspect that sometimes when it's polling at the same time, some of the polling might time-out...

maybe there's a way to set lower-priority to my apache webserver? (or is that php?)

edit : there's no silly tcp-connection limitting thingee in win2003 server is there?...

TPC · Post by **TPC** » Sat Oct 09, 2004 12:39 pm

Okay.. there's definitely something wrong on my system

Polling about 1200+ objects... it's fine on its own.
but when I log in via the web and view some of the graphs.. the polling will go awry...

any ideas?..

edit : might i add that the cpu processes isn't always 100%... it's spiky but it's average i would say is about 30-40% ?.. plus, if the polling is done concurrently one after another ( i set the processes to 1 ) - i'm sure serving one http connection with a few graph shouldn't matter that much right?...

silicon_freak · Post by **silicon_freak** » Sat Oct 09, 2004 2:54 pm

On a PII 266 HP Netserver

I am pulling 177 Objects from 29 hosts in about 120 seconds using 5 processes of CMD.

-190MB Ram
-3Com 3C905tx NIC
-SCSI 3 10k HD
-Cacti Version 0.8.6b
-Fedore Core 2
-Latest Perl, PHP, MySQL

http://alphawatch.alink.com/graph_view. ... &tree_id=2

TPC · Post by **TPC** » Sat Oct 09, 2004 4:29 pm

okay.. maybe increasing the processes count could help.. i couldn't get cactid to work with my custom script

for 1200 objects.. how high should i set mine to ?

Post by **TheWitness** » Sat Oct 09, 2004 6:43 pm

TPC,

There is a new version of Cactid (0.8.6c) available in CVS that should help with your script. I have increased the buffers for both input and output.

TheWitness

TPC · Post by **TPC** » Sun Oct 10, 2004 1:50 am

Thanks TheWitness, I'll try it out and let you know if it solves my problem. I am just hoping it's not some dinky windows tcp connection limitting thingee though, cause that seems so much like it. Whenever I am connecting from a far away place and loading the graph seems to take awhile, the polling will definitely go awry while apache's serving me the images....

Quick one, what files are absolutely essential if I want to keep my polling database and settings ... i assume cacti.sql and all the stuff inside the rra directory and my own scripts are the only things that's important right? Am i missing anything else ? I'd hate to upgrade to lose 1200 objects

I'm also thinking about making a script to upload all my data to another PC to serve dedicated webpages(for some reason the ftp export-feature is giving me a 20kbps when using php ftp upload function on a local lan-gonna do some local speed test too to see if that's a network issue). I'm gonna copy the cacti directory across to another pc every few minutes, but only this time, don't schedule a polling program on that PC but just serve files.. (that's sorta like my last resort..)

cacti is great stuff.. i'm totally addicted..heh.

edit : okay i see 0.8.6c in the cvs but i am new to all these, don't quite know what to do with the .c files..hehehe.. i'll just wait for you guys to hit me up with a 0.8.6c.zip

edit again : didn't realise about the host availability settings.. gonna try some different ping/snmp/ping&snmp settings to see if it works better

Post by **TheWitness** » Sun Oct 10, 2004 8:03 am

Here is the latest version pre-built. Don't forget to install Cygwin.

TheWitness

addlema · Post by **addlema** » Mon Oct 11, 2004 11:50 am

Specs:

Dual 3.0ghz Xeons.. 4 gigs memory....

SYSTEM STATS: Time: 78.9642 s, Method: cmd.php, Processes: 20, Threads: N/A, Hosts: 67, Hosts/Process: 4

Adding more every day.. Probably have it up to 2500 soon and 100 hosts.

Post by **TheWitness** » Mon Oct 11, 2004 12:09 pm

20 concurrent processes may be a little high. All based upon CPU. If you keep it to about 80% on both while polling with spikes to 100% you should be Ok. However, 20 does seem excessive. You should play with this setting to determine the optimal one.

TheWitness

Hitting the ceiling... Post your stats here! :)

Hitting the ceiling... Post your stats here! :)

Who is online