It really does feel hopeless to run an open-source service that isn't running on an existing crowd. Perhaps that's one reason Snowflake is doing well.

Snowflake actually has historically run as a multinenant service with access to say your S3 bucket, so the majority of data transfer is VM <=> S3 in the same Region (so included/free).

As a snowflake customer, you would submit queries (and data ingestion) as Egress pricing (roughly $.01/GB though assuming it was using the private link or similar) and then the smallish results come back as Ingress (which AWS charges for, we do not, but whatever).

I hope someone corrects me on the AWS side (I really thought that there was a way to do zero rating for same “AZ” third-party services), but it’s not fundamental.

That cost acts as a barrier if AWS launches its own Snowflake competitor. Basically AWS taxes you for not using other AWS services, once you are inside AWS.

Just to clarify, AWS has a feature called PrivateLink [1] that charges $.01/GB transferred between a Service provider and the end customer (I thought there was a free variant for same AZ, but I seem to be wrong about that).

We offer a similar feature called VPC Peering, and in particular Private Service Access [2]. Because we can be sure that the traffic is in the same Zone (or not), talking to a third-party service in the same Zone does not incur egress charges.

But did you mean this $.01/GB or were you thinking of VM <=> Internet / External IP pricing?

[1] https://aws.amazon.com/privatelink/

[2] https://cloud.google.com/vpc/docs/configure-private-services...

Super tricky and not applicable in all circumstances. GCP and AWS must move towards sane traffic pricing (or risk anti-monopoly lawsuits). The licensing changes are just a first step.

P.S. : offtopic... can i use privatelink between AWS and Google Cloud ? do you have an idea on how much it would cost on either side ?

You basically have to hookup Cloud Provider A through a middle layer to Cloud Provider B (e.g., connect via Equinix or Megaport). It’s not drastically different than a hybrid setup, except you don’t control the premises on either side.

While getting you the right link to our Partner Interconnect docs, it turns out we have a nice solution guide writing up the various options [1].

The costs boil down to some non-trivial fixed price each month to cover the 1/10/100+ Gbps connection on each side, plus whatever transfer fees each provider charges for egress over that method. At large enough volumes, providers of all sorts usually have equivalent fixed prices. Getting multiple redundant lines, or doing VPN encryption, obviously costs more and is more work than “just a pipe”.

Even at list prices though, this gets you closer to $.02/GB each way in Europe or North America. At high volume in the same metro area (like Northern Virginia), you can easily agree to much lower rates on either side.

[1] https://cloud.google.com/solutions/patterns-for-connecting-o...

> If the VPCs in the VPC peering connection are within the same region, the charges for transferring data within the VPC peering connection are the same as the charges for transferring data across Availability Zones. If the VPCs are in different regions, inter-region data transfer costs apply.

There is no additional cost

I checked with a couple of people and they read the above to mean that any peering data transfer costs the SAME as if you were transferring data across availability zones. This is not free.

Inter region transfer is a different tier