[shrubble]

It's very interesting that Dan discusses the issues of cloud latency vs on-prem with on-prem being faster; something that some on HN have pooh-poohed as not being something that mattered.

And that bandwidth costs they were paying suprised GCP , when it's been pointed out that bandwidth costs can be quite low vs the cloud providers. Granted Twitter is operating at a very large scale...

[wmf]

SLOs for most SaaS are really loose (if they exist at all). Almost the whole industry is leaving 10x-100x performance on the table because I guess very few people even notice.