TheMadScientist - 8:03 pm on Apr 6, 2010 (gmt 0)
Each and every infrastructure change Google has gone through there has been many sites damaged from the change. So if the algo doesn't change there shouldn't be any discussion on the infrastructure change assoicated with Caffeine, but since (as we already see sites are being damaged) there seems to be something changing within the algo to determine a sites ranking position.
I see what you're saying and understand the reasoning, but... A change to the underlying data or the way the data is organized can change which pages have the calculations performed on them at a given time, pages which were included in previous calculations may not be included in the process (yet or are included faster), because data may have been dropped or missed from the new dataset (or there may be more data added in), the calculations could be performed at a faster rate causing pages to be removed from (or added to) the dataset (or index, results) sooner than it would have been previously and that could be what is being seen...
There's all different reasons why the results might change from the underlying data and storage system being changed that has nothing to do with the actual algorithm (heuristic).
Here's one example (there are a bunch of possibilities):
Calculations are performed much faster (Caffeine).
A site is detected by one of the ranking mechanisms as spam.
The pages from that site have their 'weight' discounted.
The sites linked to from that site have their 'weight' updated (with the lower value passed to them).
The site detected as spam (all of its' pages) and the site(s) (pages) it was giving weight to drop in the rankings.
Using a slower system it may have taken 2 weeks to make it all the way through the calculations and have an effect on all the pages involved, but the faster system makes it happen in a day, so it could appear there has been an algo change because sites (pages) start 'dropping like flies' and what would have been 'spread out' previously 'happens at once'. The real change is not in the algo, but in how fast the processing happens, how fast a 'scheme' is discovered, how fast the removal of the scheme is applied to other sites and pages, and how fast those changes are visible in the results.
An infrastructure change can definitely lead to changes in the results without the actual algo (what's important to rank) changing a bit, by speeding up the detection and removal of 'spam' pages . The converse is also true and positive changes could make their way through the system faster. The apparent change would be to the algo, but in reality it could easily be the speed of the processing system.