Hacker News Reader

Comment by z3t4 | Hacker News Reader

How easy is it to migrate to or away from a cloud service like AWS ? I think the end-game for these cloud giants is to lock you in, so that you simply can't switch to a commodity provider.

View on HN

Despite the dreams of techies, companies rarely change their underlying infrastructure. You’re always locked in to your infrastructure choices without a lot of pain and the upside is usually not worth it.

No, the company you work for is not going to get rid of their million dollar Oracle installation either and move to Postgres just because you told the CTO that you used the Repositiry Pattern and abstracted your database logic from the underlying database.

Depends if you are using the value add services. AWS is much larger than EC2 and S3, but if you only stick to those it is pretty easy to move around.

When we embarked on our migration into AWS, we thought about this pretty carefully. There wasn't anything that we couldn't move in a year. That provides a sufficient backstop (IMO) to anything becoming untenable in the AWS ecosystem. Most everything could be moved in a handful of weeks of effort, or at least enough parts to achieve the reason to move.

AWS is still in a battle for market share; I think they're extremely "friendly" to their customers at this phase of the game. Will that change in a half-decade or decade? Maybe, but I think they're earning their profits now and likely to continue to do so.

At scale you need to move out of the cloud. Once you hit a certain point there is no economic gain.

That is almost surely true at extreme scale. The interesting question is whether that's true for 0.1%, 1%, or 5% of companies. (I think it's on the lower end of that range somewhere, provided the 5% scale cases are willing to be smart about their hybrid strategy.)

Was at a startup around 2k nodes in AWS. Bill was over 400K on average a month.

When purchasing RI I saw 800K a month.

> When purchasing RI I saw 800K a month.

I don't understand. Can you please clarify?

When buying reserved instances ( 3 year reserved ) the normal 400k spend was doubled to 800 thousand that month

Ah, thanks. What was the monthly spend after the RI purchase?

I believe Netflix and Apple use AWS, so I’m not so sure about that.

Netflix runs their application on AWS.

They do not stream out of AWS (because AWS bandwidth is too expensive for that application).

It's not only about cost. Networking is complex and especially for streaming companies.

1.) They need to know which pipes (to which carriers) are used at what percentage

2.) Be able to upgrade those pipes as they please

3.) Be able to cache content locally with those carriers/ISPs

All of those 3 points are not really possible with AWS.

Tell that to Netflix....

This is wrong, most companies at scale use the cloud nowdays, the operational cost + expertise you need to replicate those cloud platforms is huge.

How you design your system will dictate how easy or hard it is to move. I've always been an advocate of making sure the design is agnostic to where it's running. Problems with moving start coming up if you've tied yourself tightly to specific services, or ways of working. That's not to say you shouldn't use those services, many are really very useful, just be aware of, and have a plan on how to move away from them (if you need to).

This of course depends on how central the app you are building is to your business and how long lifespan you expect. If it's your core product and you expect a lifespan of > 5 years, not getting tied in with a single cloud vendor is smart. Usually that implies some kind of stack where the highest level of abstraction is a set of containers or VMs.

But when for example doing software consulting for a company whose core product and business is not software I personally advocate the completely opposite approach: embrace your cloud platform and use the services with biggest value-add (and coincidentally lock-in). PaaS-style services enable very small teams to deliver functionality very quickly and in a robust manner - and managed services allows for keeping them running very cost-efficiently.

Congratulations, now you have the worse of all worlds. You’re paying more for hosting VMs than you would for baremetal at a colo, you’re not saving money by letting the cloud provider do the “undifferentiated heavy lifting” and your developers aren’t moving faster by depending on the cloud providers services.

Reply on news.ycombinator.com