[otterley]

That doesn't sound right. Even the newest NVMe devices can't do 1M writes per second; they're maxing out at around 330k IOPS.

The 1M "TPS" you're referring to is a read-only benchmark (e.g. http://akorotkov.github.io/blog/2016/05/09/scalability-towar...). Those are reads (most likely from the buffer cache), not writes or transactions in any real sense.

[qaq]

330K iops for a single device you are very unlikely to be running a single device. There are Fusion IO models that can do 1M IOPS but they are on the exotic side. If you are optimising for throughput you can configure commit_delay so you will fsync multiple commits.

[evanweaver]

FaunaDB is faster in an unclustered, unreplicated configuration too, but that's not what FaunaDB is really for.

We should track disk IOPS, though, so you can do an apples-to-apples comparison of how much low-level throughput the database is driving. I believe the instance store disks in the EC2 C3 hardware class can support about ~20K write IOPS each.

[chillydawg]

Relaxing disk commits is a quick route to data loss. Might as well use mongo at that point.

[qaq]

commit_delay does not relax anything it's just increases latency to group multiple commits