Gaps on all my Graphs - Help

[sbadger]

On my dual 3.8 Gig Intels, I get a ton of items waitng and no proccesor utilization. I turned up the pollers and suckers with no increase in performance. I found that many of the connections were just sitting idle with no database conncection.

[Alexei]

At the moment we're monitoring about 8576 itmes on 16 hosts (linked via template). The moment I put in more hosts, we start getting gaps in our graphs AND in our data-collection.

When monitoring 536 items per host, the agent may become a bottleneck. Try to increase number of Agent threads, StartAgents.

Also you may suffer from slow user parameters, if you use any.

[drathi81]

Hi all,

thanks for your replies.
I should have mentioned that I only use SNMP-Hosts on my zabbix-machine but with 30 started pollers as mentiond by stever

I had a similar problem. Try tweaking the StartPollers and StartTrappers parameters in zabbix_server.conf. I have a very fast machine (dual opteron server), and saw no load on the server, but my queue had a ton of 5 minute + entries. Now with 30 pollers (probably overkill), everything happens instantly. Again, the Monitoring -> Queue tab should tell you what your problem is.

we're now able to monitor 13018 items on 23 hosts without any problems (except one Trunk, but this seems to be not zabbix-related cause I've never mananged to get this one working...)
Zabbix is doing it's job quite well, but it seems as we'll get trouble adding more hosts by the mysqld which seems to have problems writing all the data to the database.

Just one notice I'd like to add:
Before I set up the number of pollers I had no items in the queue with >5mintues, only itmes with 1 minute and 5 minutes. After adding more pllers the number of items staying in 5 minutes decreased slightly. Then I added some more hosts and now I've still got items in the >5 minutes queue but I do not have any gaps in my graphs...


Regards,
drathi81

[qix]

Somehow it is a performance issue.
See this post by me as well:

http://www.zabbix.com/forum/showthread.php?t=4422

It seems that database tweaking also had a positive effect.

[robbertwethmar]

i found gaps in a graph today. Checked the Load (around 1.4) and idle time (mostly 90% or more, sometimes zero). When i closed the browser that was refreshing the graphs (pushing cpu utilisation to 100%) everything worked fine with a Load of 0.8.

Why the load is to high and cpu utilisation so low i don't understand. But the machine is obviously to busy to get the data (gaps are in the data as well). If i relax it a bit by closing the browser so it doesn't have to generate graphs it works fine.

Memory usage doesn't seem to be the problem eighter. Database is 200mb, memory 1gb. But mysql is working very hard: 'select count(*) from history' took over 12 secs, but that was the same whithout zabbix running. I need a faster server than this Debian sunfire120 ;-)

[Alexei]

iBut mysql is working very hard: 'select count(*) from history' took over 12 secs, but that was the same whithout zabbix running. I need a faster server than this Debian sunfire120 ;-)

Do you run the latest 1.1.6? The select statement was removed from ZABBIX 1.14 or 1.1.5 GUI due to high unefficiency under MySQL InnoDB and PostgreSQL.

[robbertwethmar]

just installed 1.1.6. Problem remains... I was doing the select manually in the mysql client btw. I also tried to bring down the nr of pollers. But that number can't be < 6. Still, there's noting in queue... I still don't understand why load is around 1 while processor is 80% idle...

[rickardp]

Housekeeper

I have 4.5GB in history table. When the housekeeper runs it will issue an SQL statement that will lockup the history table. Switching the table from MyISAM to InnoDB made a difference (though I still have gaps but not as many).

Since I wanted to keep my history it took me about 24hours to dump/convert/restore history. Otherwise I would just truncate table and convert.

My 2 cents in this discussion

[cpicton]

Why the load is to high and cpu utilisation so low i don't understand. But the machine is obviously to busy to get the data (gaps are in the data as well).

Load average is pushed up by anything causing processes to run slowly. This could be any of:

CPU
Network
Disk IO

Your problem is probably Disk IO

[tronite]

I think you will find there are no gaps in the data. There wasn't when my graphs had the same problem.

- Paul

Does this problem perhaps re-appaer elsewhere where you'd use similar graphs, have you tried to change your physical display, perhaps connect with a different monitor?