Hacker News Reader

Comment by tzaman | Hacker News Reader

I can confirm that first hand; We're just moving our whole stack from AWS to GC and right now we're running everything in parallel with roughly the same amount of resources. AWS monthly bill: ~$1000, GC monthly bill: ~$600

View on HN

How many months will you need to run in GC to break even on the engineer time spent moving at $400/month savings?

Assuming a $400/month engineer, 1 month.

How many servers did you have to move?

I'd assume that you can count them on your fingers if the bill is only $1000/month. That's a "reasonably" small scale migration, if not locked in on AWS services.

Agreed, we have 10 instances on AWS, but it's hard to judge the complexity of the migration just based on that because these servers vary in size, requirements, and purpose.

But all in all, yes, it is a fairly small scale migration.

Generally speaking, the number of servers has very little to do with complexity of migration or existential threat from downtime, relative to company size.

Generally speaking, the numbers of server is a good indicator of the amount of services, which has much to do with the complexity of a migration.

Servers used to be beautiful snowflakes. That's not so true anymore, especially at EC2 or GCE.

Yep, we spend around $45-$50k/month and are quite happy with our support from AWS.

I would say around 8, but then again, the goal was to use Kubernetes (which we do on GKE), so much of the time was spent configuring it, and we'd use at least some of that time even if we stayed on AWS's container service.

[EDIT] Plus I'm the technical founder, so enjoy this stuff, which means a lot of the work was done in my free time.

There is no such thing as "free time".

You're just missing it. Take a walk around the block some time. Time is more than free when it is your own, especially if you're learning something new or enjoying yourself.

I disagree. Taking a walk around the block is time spent doing something that could be spent otherwise. Everything in life has a cost associated with it. I am not saying a walk around the block or spending extra time at work are bad decision. My point is no time is free, you are always trading one thing for another. Going to the gym, you could be studying. Studying, you could be growing your business. Growing your business you could be spending time with family. Its all a balance in life but to claim your time is free is naive of reality.

edit: To those down voting you are welcome to present a different perspective. A down vote by itself doesn't mean much but I welcome criticism and feedback on my point of view.

I'll bite. I didn't downvote, but I feel that your thinking is relatively binary and doesn't take the dynamism of humans into account. I can throw study material onto a playlist and hit the gym. I can go to my kid's dance recital and hit up the other bored parents there with business prospects. I can turn over programming ideas while I'm taking a shower.

In many cases, taking a break from the task at hand has great effects on resetting the fatigue that builds up when you focus too hard on a narrow set of problems. I wouldn't consider those breaks a "cost" when they're ultimately used to allow you to more efficiently use your time while at the same time reducing burnout.

Don't get me wrong - time is finite and it shouldn't be squandered. But at the same time, perhaps what looks like a walk around the block is much more than a 15 minute "cost".

Yes it is a balance, but it's important that we always be mindful not to get caught up in the rat race. Life is a beautiful thing in of itself and shouldn't be fractionalized to achieve maximum ego-gain, nor should it be entirely mismanaged that it dishevels your ego completely.

Your time is free and will continue to be so. As a human, hey you got some time left and if you got goals, well then perhaps you haven't time to waste. Yet, if you wish to speak of "reality", you must tell me who's the one demanding any payment for any second of your time? Is it no one but yourself? Is it not your idea of what that time could have (perhaps _should_ have)? No, that's not reality -- that's your superego, which is in your own head.

You should watch your tongue a little more carefully; it's rude to reply "no time is free" when it's not at all the topic of conversation. You deserved to get downvoted for that reply and the reason you're getting downvoted for this one is that it's too binary. Lighten up. All someone said is "I do this in my free time" and you had to step in as protector of truth and say "no such thing." There's no valor in this behavior.

The original was about how much cost is involved in something and what the break even point is. toomuchtodo asked a financially specific question. The reply to that question was vague and added that part of this cost was done in "free time".

My reply was perfectly suited for that response. It wasn't binary, it was begging the OP who said he/she works in his free time to stop calling it free time and answer the original question.

If you always require a short-term win, you will lose in the long-term. When the company is 100x the size it is today, the savings are quite substantial.

Odds are low almost all startups will be 100x the size they are today. Most fail. Don't optimize prematurely.

True, but if you DO hit that magic growth factor, it'll probably be too late to make a change of such magnitude. Better to get optimization experimentation out of the way (relatively) early.

For nearly all businesses the cost of service hosting is going to be similar to the amount spent on snacks. This seems about as useful as optimizing the brand of cereal you buy.

...which you should, for it's the quality of life that counts :)

That's a great problem to have.

Pro: Can't make hot cakes fast enough.

Con: I wish I could switch ovens; a white one matches my kitchen's decor.

Just curious, what type of AWS instances you use? And also is it on-demand instance or reserved instances? I'm using AWS for couple of years, exploring other options.

On demand: m1.medium mostly for Rails and Node apps and db.m3.medium for RDS (Postgres).

Now on GCE, we use Kubernetes, which means the type of instances is rather insignificant as they can be instantly scaled up when more resources are needed for containers.

thanks for the details.

m1.medium? That's two generations ago. Just by switching to m3.mediums, you'd get 23% savings. If you could see fit to switch to t2.medium, you'd be saving 40%.

source: http://www.ec2instances.info/

Seriously this is worse than premature optimization, I like to call it developer frugalness optimization. A $600 savings is not even worth the effort.

For one month, likely. How many months are they going to be running? Will they scale up? When do you decide the investment is worthwhile?

why is anyone downvoting this?, comment author has a point.

Agreed - primary motivation weren't savings, but much smoother UX with Kubernetes on Google. We're adopting container-based approach to development (and DevOps), so we explored alternatives, and chose Google Cloud.

Thanks for sharing your experience. Why, acc to you, is Kubernetes on AWS a non-starter? Is it too difficult to setup? Or maintain? Or simply isn't the first class citizen, like it's on GCP?

Or does it just make sense to stick with GCP since K8s has Google's blessing?

Or...

Two reasons really, the first is that AWS has become this bloated mess of "stuff", making it increasingly harder to use, because so much time is needed to either constantly look at it to see what's new/changed or vigilantly document everything. GKE, based on my experience, uses more "convention over configuration" approach, so the UI is much less cluttered, easier to understand, and comes with good defaults out of the box. And yes, K8s is a first class citizen on GKE and it overall fits nicely in the google compute engine environment.

The only downside is that Google Cloud doesn't have a hosted DB offering for PostgreSQL, like AWS RDS, so it took me a while to set up everything properly.

Finally, from a purely subjective point of view, Google's Material design is easy on the eyes.

This is roughly what we ended up with for our stack: https://cl.ly/1z141g0e1w38 (The top three instances are K8S, then two GlusterFS instances which hold persistent volumes for pods and finally three PSQL instances that also run Redis Sentinels with a quorum of 2 - Redis itself is on Kubernetes as a DaemonSet)

Thanks a lot for taking time to respond. I've a couple of clarifying questions:

> Two reasons really, the first is that AWS has become this bloated mess of "stuff", making it increasingly harder to use

Are you referring to any specific offerings: CodeDeploy/Beanstalk/EC2? Or generally the entirety of AWS catalog? I agree that the sheer amount of configurations and the breath of offerings might appear bloated and there are parts where AWS looks its age, not necessarily a bad thing, though.

> because so much time is needed to either constantly look at it to see what's new/changed or vigilantly document everything

I am not able to relate to this. AWS is pretty serious abt backwards compatibility and making transitions smooth unless there's a serious security risk.

Re: Console: This complaint comes up often on HN. Thanks for pointing it out.

Re: Convention over configuration: K8s seems to be a great piece of software from what I keep reading abt it. I can understand why anyone would choose to use it. I am left wondering why it isn't as easily usable on AWS infrastructure... I guess I must try it out myself, someday.

I'm not referring to any specific offering, but just the sheer amount of ways to accomplish things on AWS - and the docs don't help, I urge you to compare (or even time) the process of following the docs in setting up AWS versus GKE clusters. You'll notice the process is much faster with GKE, especially because the docs include at least some real-world examples, compared to AWS where the documentation is almost completely abstract.

And as a "very" curious developer, I always get pulled into analysis paralysis. Not so with GKE. There's one way to accomplish a particular thing, the only choice is UI versus CLI - and since most of us have Google accounts anyway, gettings started with GKE is maybe a couple of commands and you have a cluster up and running.

Try googling how to set up Jenkins/WordPress on AWS/GKE. A VERY real world examples and Google provides docs, AWS does not - I'm not interested in high-level overviews, I want to solve a particular problem.

What AWS needs first and foremost is a competent UX team.

Is something like this useful for you? https://aws.amazon.com/getting-started/tutorials/

Nope, but it illustrates my points perfectly. A company like Amazon should have hundreds of those, with real use cases that developers (or DevOps) care about. For instance, I have a Rails/Node/Go/whatever app. How do I deploy it using any of the available services? Which one is preferred? Why? What about the connecting services, like Postgres+Redis? (I know there's an offering for everything, I'd like to know how it all fits together, best practices, etc)

Makes sense. Thanks once again. :)

Thanks for the feedback on Cloud SQL! Will pass it on to the team.

I am a Google Cloud PM, though not on Cloud SQL itself.

They talked about moving to Kubernetes, so the AWS vs GCP part of this is much easier to deal with.

Reply on news.ycombinator.com