local segmentation

If they are planning to expand very soon, then enabling local segmentation may be a good idea, but if this is not happening soon, then don't enable it. I don't believe we are recommending using this anymore.

We need to qualify "very soon". Local segmentation can be nice when expanding the size of a cluster, however it's not at all required. And while I agree that a local segmentation value of 3 might be ideal in this situation, that's going to multiply all their existing projections by 3, which has a good chance of creating ROS pushback errors.

The other thing to consider here is that you can't force a rebalance just due to the implementation of local segmentation. There's no way to actually subdivide the projections into local segments just because you enable the feature. It doesn't initiate any kind of rebalance, and the system will just have to take it's own time to create those local segments. So, if you plan on upgrading in a couple of days, turning on local segmentation will have no benefit whatsoever. If you plan on upgrading in 2-3 months, then it can have some benefit. You just have to handle the risks.

as we understand, the current plan is to use 20 nodes for about 1 Pb of data and to add 10 more nodes in about a year for another 0.5 Pb. if we do not use the local segmentation, is re-segmentation of a 100 Tb a real option?!

100 Tb is the estimation of a fact table size. they will have several of them. they will be partitioned somehow by date, if it is important.