Comment by kuschku | Hacker News Reader

Comment by kuschku | Hacker News Reader

Parent Full thread

> If you need your LB to handle 10s of millions of HTTP/S requests per minute, then what other choices do you have other than Google?

The same options every business that needed them has done for decades? Renting dedicated servers én masse, colocating, or running own datacenters? And doing this in multiple countries?

What I'd suggest for a start is renting dedicated servers or colocating in a few places (EU-Central, US east, US west), and running on that.

Dedicated servers, even with having to maintain them, are always cheaper than GCP. Even more so is the network traffic.

I've started with dedicated servers with a few European hosters myself where I'd be able to get servers and traffic for 40€ a month that would have cost me upwards of 14'000€ a month with GCP, AWS or Azure (180TB+ traffic per month, I don't have the exact numbers again, but I used each service's cost calculator for this). Sure, if I want high quality networking, I'd have to pay another 360€ per month extra with these hosters, but that's still cheap enough that it means I can basically run a 180TB/month serving service on a hobbyists budget.

Dedicated servers == constanct cost. Many cases, you reach peak times for several hours per week/day. With GCP scale up and super fast, and when done, tearing down and saving money.

Also, from my experience, when you reach a certain level of operations, you discuss with Google sales team and you get a significant discount for bandwidth.

We're well past that point (with AWS) and we get a meaningful discount, but it's in no way competitive on a per-TB or per-Gbps basis with a dedicated colo.

It's also higher quality of service IMO. When AWS gets their daily DDoS traffic, they manage it and we never notice nor care. With smaller colo providers, they are much cheaper, but if your neighbor is getting DDoS'd, you notice it. I assume Azure and GCP are similar.

With regards to DDoS: that's something where OVH has surprised me well in the past, I've been a target once or twice, but they automatically handled all the upstream issues, and thanks to their hardware firewalls, I could filter the traffic before it hit my server.

On Twitter, the OVH CEO claimed that they've previously had servers targeted with 1.1Tbps DDoS, and still managed to keep that user's site working.

i assume ovh and hetzner? care to share more of your experience with them?

I've used Online (the company that also ownes scaleway, but their business line), Hetzner and OVH before.

Hetzner and OVH have much better peering than online (ofc still worse than GCP), especially noticeable when users are in Asia Pacific and you have only servers in the west.

Online is significantly cheaper, but well, the peering issue.

Generally I prefer Hetzner for compute-heavy workloads (especially their ML server offer with an i7-6700 and GTX1080 for 99€ is awesome) or networking, for most of the rest, I'd just go with Online.

Important for me is that I can just rent a handful servers, try out building something, fail, and try again, always knowing that it won't cost the world, and allowing me to experiment. If it fails, I can just try something else, if it works, I can just scale up. Doesn't have to be perfect, just has to be enough to let me fail early and often.

Reply on news.ycombinator.com