Content Filtering Trial - Results

[NetworkAdmin]

This thread is locked for easier reading and access to results. Please use this thread for discussion and comments:

http://forum.exetel.com.au/viewtopic.php?f=288&t=31837

and this thread to report results/problems:

http://forum.exetel.com.au/viewtopic.php?f=4&t=31829

[NetworkAdmin]

The filter trial was scheduled to go live Tuesday afternoon, but has been delayed because the vendor has reported that the passive filter list currently in place has recorded many more hits than they expected. They are investigating the possible causes and so we wont activate the filter trial until the cause is identified. We expect that will be sometime late Wednesday morning.

An early 'result' is four people have reported slow Internet 'due to the filter'. However, since the filter isn't active, it can be ruled out as a possible cause. In the two cases I have investigated so far, one has been due to a misconfiguration modem preventing access to any site. The second was a customer blocked for spam and mistaking the spam block for the content filter block.

[NetworkAdmin]

This is the traffic graph for the passive filter list (that is; traffic that would have gone to the filter server if the list was active) for the last 24 hours:



You can see that, apart from the peak, the traffic is mostly less than 30kbps.

According to the vendor, the traffic peaks and traffic level between 12 and 1am are unusual and they want to determine what may be the cause before switching to active mode.

[NetworkAdmin]

This is the CPU load for the router handling the filter server BGP session:



There is no change in CPU load over the week, before and after the filter bgp session was enabled. The two spikes of CPU load were caused by carrier related interruptions on the LNS routers (DSL circuits disconnected and then reconnecting) and are not related to the filter server.

[NetworkAdmin]

The filter is now active. This is the web page that will be served if a site on the filter list is accessed:

http://blocked.exetel.com.au/censored/

[NetworkAdmin]

This the an update of the graph showing traffic to/from the filter server:



The current bandwidth use since the filter went active at 2:15pm - 13.2kbps, can be fully accounted for as BGP protocol overhead between the router and server.

[NetworkAdmin]

Overnight 11 reports were received of poor or degraded performance which was attributed by the end users reporting the issues to the filter trial.

On investigation, all the reports were for either sites not contained in the filter list, or reports of general degradation of performance that could not be related to the filter trial.

A blind test was conducted to determine if there was any possibility the filter trial could be affecting user traffic.

Between 11pm last night and 9:45 am this morning the filter was disconnected from the Exetel network. Over that time seven more reports of degraded service attributed to the filter trial were retrieved.

We consider this conclusive evidence that the reports were 'false negatives'.

This is the graph of traffic over night with the filter disabled during the test period:

[NetworkAdmin]

The filter trial has ended at 12:15pm today.

This is the result summary:

• 20,000 active hits against the filter list
• no false positives
• no measurable impact on any Exetel router
• no impact on Internet access to non blocked sites for any Exetel user
• there were 56 problems reported that users attributed to the filter trial, each on investigation was a 'harbour tunnel' effect
• three people threatened to cancel their Exetel service(s) because of the trial

We were able to determine:

• how we could implement a mandatory non user optional filter system with very little/no disruption
• how we could implement a mandatory user optional filter system with very little/no disruption
• that we could offer a 'clean feed' filter system now, as a user pays option (similar to the spam filter) for a price of $5 per month per user
• the cost of a mandatory, non optional filter system would be in the order of $6 per year per user