The two tables moved are: snmp_counter64* and poller_output. Both these tables can afford to loose the data on restart – they hold data before it is moved into the rra and holds at most 5 mins worth of data (IE one cycle).

It also improves the situation if the server restarts as if data is left in the poller_output table it can push the the php poller over its memory limit and stop it working properly, so its a good idea to start with it empty.

Snmp_counter64 also needs emptying every now and again – as when you remove devices from cacti it does not get removed from here

The SQL code to achieve this is:

alter table cacti.snmp_counter64 engine=heap;
delete from cacti.poller_output;
ALTER TABLE cacti.poller_output MODIFY COLUMN output VARCHAR(255);
alter table cacti.poller_output engine=heap;


Warning
Some graphs may stop updating if the script used to generate data for the output more the 255 characters If you use MYSQL 5, then you can use something like VARCHAR(1024) instead.
This may cause data to be lost from the current run.

The server now copes pretty well. I haven’t tried anything new to it yet, but it generally appears to be working better then it has done for past couple of months. I would like to re do the disk structure, but for now this will do.

*This table is from a patch – most people won’t have this table.

Currently it has:
Hosts:354 Data-Sources:16943 RRDs Processed:8744

And has hit a limit with Disk i/o.

What I plan to do is: Move it onto a temp virtual server. Completely rebuild the old dell with two RAID 1 sets each on 2 disks – One for the RRA and one for the DB and see how it put up with that. The rebuild should help on the disk layout too – due to all the network changes and low disk space its likely the disk could have a few few frags.

If this doesn’t work and it still struggles, then I’ll use one of the various other spare dells (might have to be one without a raid card – so it’ll be software raid) and stick the database and possibly front end on it.

Once we have it stable again – I suppose I better look into a better way of coping with gigabit Ethernet cards in windows – currently we use the SnmpProxy2.pl hack – may just be a case of limiting it to Windows servers some how and not all 64bit Ethernet counters.

A bit of messing around – including installing the version from Testing I found out I need a new kernel for WPA.

Of I went and got 2.6.17 and compiled it.

Things of Note:
I had to use the ieee80211 module and the ipw2200 module from the kernel – but they appear to work so now problems.

After that network manager poped up – it found my wireless network. Selected it, a setup box appeared in which I put my WPA pass-phrase in and off it went and connected – all working a lot more reliable that just wpa supplicant on its own. Could be down to the new kernel mind. However it is very easy to use, and well I impressed. It’ll make using my laptop as a laptop a lot easier.

Update
Wow – after all this work and finally wireless is the most stable I’ve ever had it on Linux. In fact its not gone wrong once since this. Even in windows I was having a few problems now and again. Speed is good, stability and great and it just works. I enter a password in at the start and it goes off, picks the right network and passpharse and just sets it up for me.

Basically allows you to draw diagrams of you’re network, and it takes data from the rrd files and displays different coloured arrows show the percentage of the link used.

Basically allowing you to see whats going on, on your network and where all the traffic is going/coming from

However at work we have started to look into Open Source.

We currently have an every-growing Cati Install. Cacti is a great bit of software – at its most basic it creates graphs from SNMP data. It makes it very easy to add new hosts and created and managed display them. This is a good thing in our network – we are currently monitoring 116 devices with it, and this is growing every week as I get around to adding more devices.

With a number of addons (see http://cactiusers.org/) Cacti can also do a lot more. Currently the system:
1) Graphs various data from servers, switches, UPS’s and everything else we feel like monitoring with SNMP
2) With a few Perl scripts it even monitors various data from the likes of Exchange, ISS and Citrix (Using nsclient which can pull data available in WMI to Linux)
3) Monitor servers for restarts (via checking uptime). Emails us when every a server restarts
4) Threshold Monitoring – the system can email use when ever a value has gone bellow or above a value for a certain amount of times (mainly used to warn about UPS usage and disk space)
5) Reporting – Can build reports from the rrd files to find out top talkers etc

We are adding more and more features to the system all the time. Its very helpful to have a system which sits there and monitors everything we throw at it so we don’t have to keep a close eye on things.

The next open source system was are looking at using is good old Squid. We are getting a number of 100mb net connections at various sites and we need to replace of aging ISA boxes. Given that we have loads of old hardware available we’re going down the squid box. If we find ones is performing slowly just build another one to help out. Currently in testing and working out what rule and routing needs setting up still.

Currently using:
Kernel 2.6.14
-Wireless Extension v19.
Debian Etch + Extra apt sources
wpa_supplicant v0.4.7
Wireless-Tools version 28

The problem this time is that I needed to change some of the options for wpa_supplicant. No longer does it work with the ipw drivers but the wext drivers

/etc/default/wpasupplicant now reads:

ENABLED=1
OPTIONS=” -D wext -ddd -i eth1 -c /etc/wpa_supplicant.conf”

It works for now, but a new kernel compile is needed as the laptop doesn’t powerdown on shutdown at present. So may be yet another upgrade needed to get wireless working again.

I have got it working, but I used a custom 2.6.12 Kernel (the debian one doesn’t quite agree with all the stuff on my laptop), and compiled and installed ieee80211-1.0.2 from source, then compiled ipw2200-1.0.6 from source and it works! Well for now. Just in case its important the version of WPA_Supplicant I’m running is 0.4.6.

Time to see how long it lasts this time……..

(Oh and by it works means that most of the time it loads in the right order and works – sometime it loads in the wrong order so it ends up been eth2 and not eth1, other times I need to remove and re-add the ipw2200 module for things to work)