Hacker News new | ask | show | jobs
by kicktheshoe 4836 days ago
This benchmark provides some interesting data, but would have been far more useful if they had provided more details. There are many unanswered questions here:

How even is a comparison when GCE hardly has any customers sharing the same physical host and network at this stage? It's rather interesting that EC2 read I/O performance was higher despite this consideration.

What will happen when you start sharing a physical host on GCE with many other customers? Will you all still get 157 MB per second disk write throughput simultaneously?

What type of instances were used (and were they EC2's first generation or second generation instances?). Given the fact that they are looking at ephemeral disk performance it's probably a 1st generation instance.

I see no mention of EC2 EBS provisioned IOPs. If performance was a goal then surely having guaranteed provisioned I/O should have been considered? Does GCE even have that feature? Surely IOPS are more important than sequential I/O for most tasks?

When benchmarking EBS I/O were EBS-optimized instances selected?

Why no mention of EC2 cross region AMI copy? Also GCE is currently in the US only. EC2 is on 5 different continents currently.

What are the costs of the instances (using reserved instance pricing for EC2)?

1 comments

> GCE is currently in the US only.

Not correct. They've had EU zones since November:

https://developers.google.com/compute/docs/faq#datacenters

https://developers.google.com/compute/docs/pricing#table

Seems the FAQ has recently been updated, a week ago there was no mention of Europe. Odd.