That doesn't actually speed things up. Generally, in fact, it slows things down.
This is a case of sequential update. Batch update converges more slowly in terms of the total number of operations, but because of the economies available in map-reduce programs (due to sequential reading, merge sorting, shared nothing and so on), the convergence in terms of time is considerably faster, especially if you hold total hardware constant. This is *exactly* the same point that I have been making. You should *not* be doing random access during a page rank computation (not if you want high speed). On Fri, Jul 3, 2009 at 2:13 PM, Marcus Herou <[email protected]>wrote: > > We use memcached during PR calculations to store the node's temporary score > so whenever you calculate the score for another node which is dependent on > the node in question you can access the previously calculated scores. Was > that explicit or just confusing ? Trying once more...
