Re:$2 H100s: How the GPU Rental Bubble Burst
by jgalt212 on Friday October 11, 2024 @07:37AM
(#41808551)
Cloudflare caching of static resources is cheap, so back to one closet. But three if you want to be pure and totally cloudless.
[ Parent
| on HN ]
Re:$2 H100s: How the GPU Rental Bubble Burst
by felixgallo on Friday October 11, 2024 @08:31AM
(#41808904)
With one closet you can also lose the entire business if one water pipe breaks or one wire goes bad in drywall. Back to three closets.
Re:$2 H100s: How the GPU Rental Bubble Burst
by llm_trw on Friday October 11, 2024 @08:56AM
(#41809070)
If only it were possible to make backups. Alas no such technology exists.
Re:$2 H100s: How the GPU Rental Bubble Burst
by dartos on Friday October 11, 2024 @09:35AM
(#41809364)
backups do not prevent downtime.
Re:$2 H100s: How the GPU Rental Bubble Burst
by krab on Friday October 11, 2024 @01:08PM
(#41811282)
Yes. That's a risk assessment every company must make. What's the probability of downtime vs the development slowdown and the operating costs of a fully redundant infrastructure?
I worked for a payments company (think credit cards). We designed the system to maintain very high availability in the payment flow. Multi-region, multi-AZ in AWS. But all other flows such as user registration, customer care or even bill settlement had to stop during that one incident when our main datacenter lost power after a testing switch. The outage lasted for three hours and it happened exactly once in five years.
In that specific case, investing into higher availability by architecting in more redundancy would not be worth it. We had more downtime caused by bad code and not well thought out deployments. But that risk equation will be different for everyone.
Re:$2 H100s: How the GPU Rental Bubble Burst
by llm_trw on Friday October 11, 2024 @07:35PM
(#41814978)
The 2018 server in my garage has had a better uptime than in the last 6 years.
Re:$2 H100s: How the GPU Rental Bubble Burst
by LoganDark on Friday October 11, 2024 @09:09AM
(#41809161)
Yes. That's what the other closets are for. Redundancy.
Re:$2 H100s: How the GPU Rental Bubble Burst
by packetlost on Friday October 11, 2024 @10:48AM
(#41809943)
Very few businesses are living and breathing by their system uptime. Sure, it's bad, but having a recovery plan and good backups (or modest multi-site redundancy, if you're really worried) is sufficient for most.