Technological dark matter

September 2015

Some tech companies seem ridiculously large proportionally to what they do.

To pick some random examples: Dropbox is 800 people, and their major product hasn’t changed since ‘09. Facebook is pushing 10,000 employees to build a website–sure, a really big website, but it did basically the same thing a decade ago. (And these examples are basically random; I could pick on most other large tech companies.) What gives? Are they just really unproductive? Do their engineers work on unimportant things? What do they do that’s big enough to justify their headcount?

I think that often software companies do over-hire and then work on silly things. But there are also a lot of things beneath the iceberg-tip app that most people come into contact with:

Enjoyed this post? Get notified of new ones via email or RSS. Or comment:

Public (as ) submit ⤇

Dan Luu

This is covered by your microoptimizations point, but it’s the opposite of your example: you can save a mind bogglingly large amount of money by building better infrastructure.

Say your costs are something like Table 4 from http://web.eecs.umich.edu/…/publica…/hauswald15djinn.pdf. Maybe each server and the attached networking gear and surrounding dataceter plus power costs you (amortized) $3k/yr, and you have 1M servers. That’s 3B a year. A tiny 1% micro-optimization is worth 30M a year.

Really large companies (including FB, to some extent) build a lot of their own hardware, including power supplies, networking equipment, physical racks for servers, boards for their machines, custom chips, etc. Going one level higher up, they also write their own custom networking stacks and protocols, do large amounts of customization to their operating systems, and so on and so forth. Going another level higher up, they also write their own filesystems, databases, machine learning libraries, mapreduce equivalents, etc. It takes thousands of engineers to do all of that stuff, but the cost savings is much larger than 1% for each. It’s a no brainer to do that stuff once you get beyond some size.

Most of these things have no obvious direct impact on a product, but doing a standout job in one of these areas can literally save a billion dollars.


Heidi Et Al

“Paypal was famously within days of being destroyed by fraud.” I am out of the loop and having trouble finding information on the internet about this. What’s the story?