Shaddows - 1:34 pm on Mar 16, 2010 (gmt 0)

Despite recent daily updates our home page cache is now 14 days old

Very often the case pre- and mid-update. The Cache date does not necessarily imply your ranking is out of date.

You might want to consider the noarchive tag though. Search the site for details, incrediBill being a strong advocate, from memory.
-------------------------

Further to WN, the fact that slightly different data is held on different datacentres necessitates different datasets.

Consider this. On an average day, you can see slightly different results on different DCs. These chatter to each other, keeping them largely aligned.

During an update, due to imperfectly matched data, each DC builds its index in slightly different ways. Small changes in initial conditions mean the index is built in different ways (priority given to differnet data shards). These converge as iteration reduces the significance of the the order of processing, and also as base data becomes more complete.

However, convergence is not absolute. To deal with this, two things happen. First, Google has predefined "important" datapoints, such as particular homepages. These act as data anchors, starting points if you will.

Secondly, there is the inter-DC chatter I mentioned earlier.

Under Caffiene, I'm expecting this folding process to disappear, as physical DCs become irrelevant, and all aspects of the Google world become distributed (and not just the loadbalancing we see today).

Disclaimer: This is a mix of fact, analysis and conjecture, brought together into a single cohesive view for the purpose of clarity. It is also overly simplistic, avoiding certain subtleties. Free forum advice is worth what you pay for it