Welcome to WebmasterWorld Guest from 188.8.131.52
Big Daddy has got to be about hardware - native 64-bit operations will allow fast access to lots more memory than normal 2 GB - it is possible to access more than that while being in 32-bit mode, but memory speed won't be that fast. Dealing with data using 64-bit registers would also help performance. All in all pretty big changes which explains why it takes so long to release it.
I dont believe that moving to 64bit hardware is going to cause any improvement to the canonical issue, and that is something Matts told us to keep an eye on for improvements.
'Infrastructure' is so wishy-washy it could mean anything, but if it was just a hardware update I doubt we would even notice...
This being the case, whilst it is possible that Google is testing a 64bit platform, it is unlikely that they would conduct such tests using prototype algos. In other words, when testing a 64bit platform, they would compare results with those of a 32bit platform and expect them to be identical.
So why would a move to new hardware change the index
Its not a secret that new index is much bigger than current one, seems to be twice as big - 16 bln pages. If it is indeed a 64-bit platform (necessary to break through 2 GB memory limit without memory bank switching penalties) with more memory and possibility to rewrite some key hotspots using 64-bit assembly, then it makes sense to increase data sizes so that this move is justified.
Why not port existing code and index to it first? Because it must have been done already and such a huge hardware change should ultimately come with real improvements visible to the end users.
>A change to 64-bit IDs in the indices is reasonably likely to be part of this revamp, but that has nothing to do with 64-bit operating systems or 64-bit CPUs.
I don't think so. Taking into account the massive number of operations needed for PR-calculation it is quite clear that this can only be done with very concise programming on machine-language-level. Given this, it is a huge difference whether you can acces the whole index with one CPU-step or whether you need two or even more.
I also think that going beyond 4 billion pages has required a massive restructuring of the algos on just this machine-language-level. On the other hand this 32/64-bit disussion now lasts for more than two years and I doubt it takes so long for google to exchange the hardware. It seems likely that there is much more to big daddy than just this issue (eg IPV6?), but indeed it might play a role. However, it'll always remain speculation.