Live data from Hacker News

Cloud Egress Costs

getdeploying.com

71–80 of 318 posts

Re: Cloud Egress Costs

#71
I wonder what would be the implications of AWS egress traffic increasing significantly (like order of magnitude). It is easy to rant about costs when you do not need to consider the consequences. Obviously AWS would need to build up their own network capacity, that's the easy part. More importantly they would need build up interchange infra too. This is the point I struggle to even imagine what the scale is like. Is it just plopping in couple of new 100G ports, or would that be just a drop in the ocean and in reality its more like couple of racks of network gear. Then there is the question how the ISPs/peers AWS is connected to are able to handle the traffic; AWS is not able to just unilaterally increase their egress capacity, the traffic needs to go somewhere. Besides technical questions, such increase might also cause renegotiations for peering agreements especially if the traffic balance shifts dramatically. And so on.

This got me also thinking, would operating some sort of spot market for traffic make sense to AWS? I imagine for them the main expence is maintaining capacity and that to me implies that optimizing off-peak use could be beneficial.

Re: Cloud Egress Costs

#72
post #59
post #48

Earlier quoted context omitted.

Your little NAS is not directly connected to hundreds of metro areas around the globe. Even on Google Fiber, which is an excellent network, you only have a few dozen points of exchange, all in America. As a Google Cloud customer your egress traffic will be on Google's network all the way to the very edges of the world, at hundreds of interconnect points and exchanges. Your traffic will be on private networks until it…

Except they will charge you the same extortionate egress fees for outbound traffic from a single compute instance ("an apple"), no global content delivery network involved ("a diamond").

It doesn’t matter where your instance is, the point is the traffic will flow on Google’s worldwide private backbone until it gets very close to where it’s going.

Re: Cloud Egress Costs

#75
post #65

Oracle Cloud only charges a fraction of want Google, Microsoft, and Amazon charge. Any idea how Oracle is able to keep the cost so low? Or are the others just inflating the price so customers don’t move to the competitor? In that case Oracle deserves a shout out for not applying these vendor lock-in practices.

Other are just charging what they are because people are too ignorant to complain/object and will just pay.

Re: Cloud Egress Costs

#76
post #46
post #25

Earlier quoted context omitted.

My feeling is that egress is easily measured, so it's where costs that are hard to assess get moved to. It doesn't feel great to be line item billed for stuff at 10x the rate of credible other offers. I think there is also some geo-specific pricing that gets hidden in a global price; bandwidth can be a lot more expensive in some locations than others and if you are charged 5x for egress in south america, nobody will…

I don’t believe this. Operating an internal cloud network is expensive , but it’s expensive because of internal traffic, and they don’t charge for that internal traffic. Egress is just like traffic to any other system, and AWS doesn’t charge for that. Also: > It doesn't feel great to be line item billed for stuff at 10x the rate of credible other offers. It’s quite a bit worse than 10x

AWS charges for internal traffic too, both cross-az and cross-region.

Re: Cloud Egress Costs

#77
post #25

Earlier quoted context omitted.

My feeling is that egress is easily measured, so it's where costs that are hard to assess get moved to. It doesn't feel great to be line item billed for stuff at 10x the rate of credible other offers. I think there is also some geo-specific pricing that gets hidden in a global price; bandwidth can be a lot more expensive in some locations than others and if you are charged 5x for egress in south america, nobody will…

Right. Egress is an imperfect, but reasonable metric for overall utilization. If they started charging for CPU hertz above a certain threshold, that'd be a harder sell.

> If they started charging for CPU hertz above a certain threshold, that'd be a harder sell.

You mean like EC2 T series burstable CPU credits?

Re: Cloud Egress Costs

#78
There are many reasons the Netflix uses AWS for everything except serving the actual video bits, but this is the main one. Even before they built their own CDN they used other providers with much cheaper egress.

Re: Cloud Egress Costs

#79

The cost of egress traffic is a very good reason for many organizations to not fully migrate to a cloud provider anytime soon. And since, unlike with storage costs, there doesn't seem to be an actual reason (other than: it makes migrating to competitors cost-prohibitive in a subset of cases), that seems kind of... weird? Small example: an actual company I do some work for is in the business of delivering creative ass…

If you still want to use some AWS services, you can get an AWS Direct Connect fiber cross connect from your data center rack to AWS, just like you do with your Internet connections. They operate Direct Connect in lots of third party carrier-neutral data centers. AWS egress over Direct Connect is $0.02/GB instead of $0.09/GB over the public Internet. You can serve customers through your unmetered Internet connections while accessing S3 (or whatever) via Direct Connect on the backend.

Re: Cloud Egress Costs

#80
post #62
post #60

Earlier quoted context omitted.

Their list price for storage capacity is only on the order of 2x what S3 charges, and Snowflake and S3 likely both offer discounts. Comparing compute costs is harder. If I were running a service like Snowflake, I would certainly appreciate the effortless scaling that the major clouds offer. But I also know what I pay for actual servers and roughly what I would pay in a major cloud, and I much prefer the former.

tbf, snowflakes whole thing & what makes them "next gen" vs a traditional datawarehouse is that you dont may much for data at rest. You pay as little as possible for the "warehousing" part, and pay mostly/only for compute/querying.

wait, you still pay your S3 storage bills, right?

just with DataBricks. you don't pay when you don't use their software, cool. but then you pay for per core per second for the license in addition to running the VMs, right?

sure S3-like storage is 'cheap', but if you want up-to-date data pipelines you'll end up running these things for many hours each day. and plus you pay for each bucket operation, and so on.

Post reply on HN