Load from disk incurs heavy latency cost each iteration. Perhaps we can reuse intermediate graphs to avoid frequent loads.
cyl
Why do we need to stick to the process of map -> reduce instead of writing some specialized program for our need?
lol
You don't have to use map-reduce. It's just saying that in this context, i.e. the pagerank graph processing algorithm, using the map-reduce framework would entail loading form disk on each iteration.
Load from disk incurs heavy latency cost each iteration. Perhaps we can reuse intermediate graphs to avoid frequent loads.
Why do we need to stick to the process of map -> reduce instead of writing some specialized program for our need?
You don't have to use map-reduce. It's just saying that in this context, i.e. the pagerank graph processing algorithm, using the map-reduce framework would entail loading form disk on each iteration.