Hacker News Reader

Most of those billions of values are not shown... why not compute only the values that are on screen?

That's exactly what we're doing :) The issue is that most of the time people look at aggregates and to compute those you have to compute all the leaf cells.

Can I ask — why do you need all the leaf cells? If we want to know total gross profit for the year — why can't we compute that as "total revenue - total COGS", rather than aggregating every net profit split by product / region / month?

Are you suggesting that this is a two step aggregation , first by groups, second, ultimately, aggregate those groups, in which case the first step was unnecessary?

In any case, if you sum a list of numbers you have to look at each value. There’s no way around it. You can parallelize, like MapReduce, but the number of items is unchanged. You can preaggregate, but at some point you need partial sums.

Reply on news.ycombinator.com