Double-Wide Rack Cooling

Monitoring downtime (it isn't about uptime)

All network management products miss the point.  Their primary focus is on monitoring uptime.  This serves no business purpose and cannot justify any return on investment.  All the development is in looking at when the situation is acceptable but none when you are in dire straits.  How is this monitoring?  All it does is give you a comfortable feeling.  With this approach there is no difference in the value proposition of 'ping' or a network management framework product worth a million bucks.

What is required is to monitor downtime.  Outages require a special view that is serialized and has the outage time.  However, it is not the length of time of the outage that is important but the crucial time periods within the outage the provide the metrics.  These metric times are aligned with the expanded incident lifecycle.

Where is the real worth in reporting 99.9% availability?  This would translate to 0.1% outages but this is meaning less because:

• do we know which services were affected
• what was the business impact
• what were the prevailing conditions during the outage
• what was the resolution (and was a workaround implemented)
• did we annotate the visual, proximate and root causes?
• do we know the resources assigned to work on the outage
• how was the outage classified and prioritized
• what was the risk assessment and mitigation

A metric of 0.1% answers none of these questions and no network management product in existence reports on them.  As an example, does a historical snapshot exist of other outages at a certain time of a major incident?

Crucially, the monitoring of downtime needs to record the delta in the time when the actual outage happened against when the outage was acknowledge and escalated to the responsible operator via a notification process.  If there was an outage with no notification then this needs to appear in an exception report.  Every network management product vendor bases their sales pitch on the fact that they will provide a faster detection time for an outage.  This is absolute bull and the reason is that in most cases, when the downtime of an outage is analyzed, the detection time is the smallest contribution to the overall length of the outage downtime.  Why would you buy a tool that's primary purpose is to address the smallest requirement?

The other misnomer is that a monitoring tool will somehow miraculously prevent an outage.  The reality is that shit happens, and if your tool is only focused on prevention it will fail when you try to use it to cure.  The value of a tool is how it deals with an outage because the reality is that this is the major factor in dealing with service improvement.  Optimal management of outages results in two trends:

• the time period between outages increases
• the length of the outage reduces

There is no value in your monitoring tool if it cannot influence these trends and worse if it cannot even report or measure them.

The next step is diagnosis.  Diagnosis monitoring is not that difficult.  What often happens is that an outage occurs; a techie diagnoses it and after an investigation comes up with a solution.  Often this knowledge remains in the techie’s head and is not annotated in a separate knowledge base.  If this was consistently done, then the diagnoses time would reduce and immediately a tool that provided this ability would have a justified return on investment.

The Repair-recovery-restore time periods are not difficult to measure if a network management product is service aware, as these metric times will be obvious.

The output required is an analysis of the consequences of downtime:

• what is the total outage time and what was the time period of degradation (brown outs versus black outs)
• what is the deviation off the norm for this outage compared against the histrical trend
• what are the average incident times for: detection, diagnosis, repair, recover and restore.
• what are the top services affected by outages
• what are the top causes of outages
• what is the MTTR, MTBF and MTBSI?
• what is the Manhattan analysis?

Network management needs to step up a gear from Red-Amber-Green, to different shades of grey.

A brand new bakkie

Do you know that when a woman wears a leather dress, a man's heart beats quicker, his throat gets dry, he gets weak in the knees, and he begins to think irrationally???

Ever wonder why? 

It's because she smells like a brand new BAKKIE!

(via amaree of KSS)

Is Shaheen Saib the scarlet pimpernel a.k.a pink ninja

BBC coverage of Apollo 13

The Daily Maverick wrote about Apollo 13, 40 years on!

Video conference: We are the World