James Downs wrote:
For Netflix (and all other similar
services) downtime is money and money is downtime. There is a
quantifiable cost for customer acquisition and a quantifiable churn
during each minute of downtime. Mature organizations actually calculate
and track this. The trick is to ensure that you have balanced the cost
of greater redundancy vs the cost of churn/customer acquisition. If you
are spending too much on redundancy, it's as big of mistake as spending
too little.
Actually, for Netflix, so long as downtime is infrequent or short
enough that users don't cancel, it actually saves them money. They're
not paying royalties for movies being streamed during downtime, but
they're still collecting their $8/month. There is no meaningful SLA
for the end user to my knowledge.
I imagine the threshold for *any* user churn based on downtime is very
high for Netflix. So long as they are "about as good as
cable/sattelite TV" in terms of uptime Netflix will do fine. You would
have to get into 98% uptime or lower before people would really start
getting irritated enough to cancel. Of course multiple short outages
would be more painful than a few longer ones from a customer's
perspective.
I imagine Netflix is mature enough to track this data as you suggest,
and that's why they use AWS - downtime isn't a big deal for their
business unless it gets really, really bad.