One thing that never gets mentioned but that seems important to me is the time factor of scaling. Say your web 2.0 disruptive foo.com starts out with whatever t

One thing that never gets mentioned but that seems important to me is the time factor of scaling. Say your web 2.0 disruptive foo.com starts out with whatever tech stack you'd like. The RDBMS upgrade path is several discrete and large increments from MySQL on a VPS through beefier hardware, read slaves, Oracle, shared storage, faster SAN, etc. Each one of these steps adds moving parts and substantial engineering time to get to the next plateau. If you're scaling predictably that's fine but for a lot of startups the possibility of having large unpredictable traffic spikes and not a lot of cash in the bank isn't exactly ridiculous. Being able to have an ops staff that knows one tech stack (say Cassandra) means you can upgrade incrementally by adding small amounts of capacity in-place vs forklift upgrades. If you get featured in $media_outlet you buy more servers which add to your capacity instead of replacing it.