Hacker News Reader

Comment by web3-is-a-scam | Hacker News Reader

This article matches my experience with RDS. Performance is just absolutey atrocious when even a docker container on my MacBook performs the same queries on the same dataset at 100x the speed.

M1 with 16gb of memory vs 16 vcpu Xeon with 128g, my laptop absolutey trounces it.

View on HN

That doesn't sound right at all.

Sounds right to me. Postgres on AWS doesn't work because relational databases don't work on networked storage.

Not really. EBS isn't really network storage in the traditional sense. It's closer to iSCSI-attached NVMe over a dedicated low-congestion storage backbone network.

Lol I get what you’re saying but it’s funny your description is essentially “storage that is networked”.

This is very simple:

EBS has a latency of around 2 ms.

SSD has a latency of around of around 0.25 ms.

A relational database will have around 10x the performance on an SSD compared to EBS because relational databases need to ensure data has been fully written to disk.

Sometimes. The more expensive varieties of EBS provide comparable latency numbers to local SSDs (by using fancy on-hypervisor volatile storage or something? No idea), for an added cost.

A separate set of behavior exists for RDS Aurora, which isn't using EBS underneath, so it might be worth looking into the Aurora latency characteristics (vs. cost, of course; Aurora ain't cheap) if you're concerned about the performance impact of EBS-vs-local-disk.

Ultimately, hosting a relational database on AWS will cost 10x as much regardless of which non-working flavour of it you choose.

Million dollar database bills do not happen outside of the cloud world.

> Million dollar database bills do not happen outside of the cloud world.

Laughs in Oracle per-CPU licensing terms

Reminds me of when a customer complained a DB-heavy processing step took about 10x as long as expected, ie 10 minutes instead of 1.

We used on-prem DB server and fat clients. After a long debugging session I hadn't found anything. Gradping at straws I asked what kind of network they ran.

Turned out the clients all ran on laptops connected through Wi-Fi... so yeah 10x latency turned into 10x longer job.

I wasn't saying anything about Postgres not working on AWS but rather the alleged speed difference.

In theory, Postgres will be 10x faster on bare metal than AWS. From practical experience that is also what I see.

It's the difference between using SSD and EBS for your disk storage.

Reply on news.ycombinator.com