I've never had a client question GCP on the basis of Google's pattern of removing services. I think people can logically separate Google's consumer-facing offerings from their paid cloud services, the same way people logically separate AWS from the Amazon store. I've never noticed a GCP service removed in the nearly 10 years I've used them. In fact, the pattern I see with GCP is that they're slower to add services , but the ones they do release are very well thought out and solid. AWS, on the other hand, seems to add tons of services, some of which are incomplete or never get traction.

I would happily use either AWS or GCP. IBM Cloud and Azure I'm far less inclined to use, although I can understand Microsoft shops wanting to use Azure. IBM Cloud is pointless.

My experience is that Google services are often an order of magnitude harder to use than competitors.

For example I did a shootout between the visual recognition APIs from IBM, Amazon, clarif.ai, Google and some others.

I had the other ones up and running in under ten minutes each.

Google made me go through a sign-up and authentication process that looked like

https://www.youtube.com/watch?v=sWEvp217Tzw

and then installing their Python SDK trashed my Python installation and forced me to reinstall anaconda.

I think Google is used to hiring people with 130+ IQs and wasting their cognitive capacity and they would love to make you waste your cognitive capacity too.

Huh, I think most people feel the exact opposite, especially with the comparison between AWS and GCP. GCP has lots of services that are really easy to get set up and running easily by a small Dev shop (Firebase is pretty fantastic in my opinion), while with AWS unless you have some DevOps/networking experts it's a lot harder to do things correctly and securely.

If anything, I think the biggest issue with GCP is they still don't have the "DNA" to do enterprise support at the level large companies expect, while AWS does.

You nailed it. Coming from 8-years of AWS and working on GCP for the last 4, enterprise support is the BIGGEST differentiator IMO between AWS and GCP. GCP's services are usually technical advanced sooner (encryption architecture, various of compliances certification, etc.) and more developer friendlier, their support is absolutely the worst in the industry, opaque, slow and hard to reach, just like the support from any other Google consumer facing products. Google has never been a consumer friendly technology company, probably never will. This is in dire contrast with Amazon, which Bezos claim it always has been a customer focused business.

What about Google support? Nada. They don't even have a direct phone number to reach (requests are sent via email). There are issues that they completely dropped the ball on and provide no updates whatsoever or reluctant to look into or to put a closure on. Day and night difference. (We weren't a $1MM account in terms of total annual GCP spending just yet, we very well could be in the next years I hope.)

https://cloud.google.com/storage/docs/interoperability

https://www.microsoft.com/developerblog/2016/05/22/access-az...

If you think I'm wrong, imagine for a moment what life would be like if every RDBMS client in existence got written in terms of Oracle's wire protocol, back when Oracle's RDBMS offering dominated its respective market. Thankfully that parallel universe is somebody else's problem right now, but it could have happened.

Your Oracle example is flawed for many reasons. SQL standard pre-dates Oracle and I am not even sure what you mean by every RDBMS client got written in Oracle's wire protocol. Ingres also predates Oracle. Are you implying that S3 API is there because they were the first? Nothing else?

https://en.wikipedia.org/wiki/Ingres_(database)#Early_histor...

SQL standard:

Developer ISO/IEC First appeared 1974; 45 years ago

Oracle first release: Oracle v2 2.3 1979

And also, it is allowed and trivial to implement S3 APIs because it is well documented and available for anybody. You are also implying that S3 API is not a "properly abstracted storage layer"?

Yes, many RDBMS offerings have preceded Oracle, but none gained such wide adoption early on. Oracle has always had a strong sales team, so they've been able to grow their business rapidly. Good for them. SQL standardization has nothing to do with it either, because SQL is the user-facing language, not a wire protocol. What travels over the wire between an Oracle client and the server has about as much to do with SQL as the S3 API has to do with XML.

In your rush to defend AWS you seem to be falling into the same trap of immaturity as the people who have assumed that S3 will always remain the one true storage layer. It is not your fault, just a sign of the times. You're part of the generally apparent downward trend in level-headed, long-term thinking among the software cadre. For your next project, I recommend that you keep in mind that proper abstractions and clearly defined interfaces are crucial for maintainability of software over the long term. Good luck!

both look like third party projects rather than something by Google/msft. saying that "gcp/azure has a s3 API" would be like saying "Linux has a win32 API" because there's WINE.

GCP do have a clause in their SLA/T&C where they promise a 1 year notice in the case of any services they plan to deprecate, but that hasn't helped assuage fears - they want longer timescales and better support.

> That's just not enough when you need something easy to setup that you can build then forget about for a decade. That's just the reality of how software projects are done for non-technical businesses

Thank you for phrasing this, a few years ago I would not have believed it, and after some experience this resonates quite well.

Why? Microsoft is known for long term support, at reasonable prices. Any business that's survived more than a decade loves them for it.

We had clients running Windows 95 in VM's if that gives you an idea of how far some companies will go with "if it ain't broke don't fix it". It wasn't an usual request for IT help keeping cousin Bob's old VBA macros going another decade. Maybe 20% of our revenue was for crazy shit like that. And for good reason, many times we gave them a quote to rebuild the system (we did about 50/50 IT/Software) and it was indeed more expensive than hacking something up to keep it limping along.

1 year support is a joke. That's not even enough time for a company that builds software to migrate.

Puh-lease! I log into a VMS server every day to access an application that saw it's last update in 1995. Which wasn't even deployed until 2001. It will be replaced this year with a Win95 era application, complete with acres of giant grey buttons, served over a Citrix link from VMs running Server 2008. Bleeding edge, boys, bleeding edge.

So, if business values stability over risk, I understand very well if they select very very very very long term support option rather than to create the product from scratch.

It's not enough to finish writing the requirements. One of our customers won't touch anything unless 15 years of support is guaranteed. And that's what they get.

Obviously there is a highly domain specific application for our Civil & Structural works last updated 10+ years ago. No one familiar/willing to upgrade it's internal licensing systems to work on newer Windows. So here we go. Whoever needs to work on it, take Remote Desktop. Simple.

But since these guys have now switched to a subscription licensing model, they no longer do that, so there are scattered XP boxen around running ancient versions of various expensive software that just keeps working.

It’s the reason we haven’t adopted the latest replacement. Why bother?

Edit: thanks, it was indeed FPS.

I was in a team that worked on "another" :-). Some of the best memories of my professional life were formed during that period! At launch that was indeed as part of AWS. However it then came under Amazon Payments entity due to legal reasons. Felt sad to see it retired though[1]

[1] https://forums.aws.amazon.com/ann.jspa?annID=3045

GCP comparably has been pretty good to me I n terms of support. But that’s probably because I actually have a support contract with them. (And, I didn’t with Amazon)

I think most people who complain are not /actually/ buying support.

The expectation being that support is baked into the service price. But this isn’t the case with any of the cloud providers as far as I understand.

They would reply. Please see documentation (same link I just sent). After the second or third try I just stopped bothering.

makes me laugh that they think 1 year notice is enough. should be at least 5 for my business to properly consider it.

If everything on a platform is an emergency, that vendor won’t be around in those businesses.

Could you give me an example of something you had trouble looking up? I am genuinely curious.

As far as experience - the Cloud hasn't lived up to the promise for me. We often use Azure function apps and debugging them is a pain. Sometimes the function doesn't load at all when you change one thing, etc.

AWS was a bit of a better experience, and I think almost everyone knows what "S3" is. That is to say that AWS immediately springs to mind when you need to store BLOBs.

I have also read about people having a lot of problems with GCPs BigQuery - infrequent, but random downtimes with support not being able to figure out the issue.

Edit: As someone else said below - I do agree that all Google's services have a clear use, whereas AWS and Azure tend to have a lot that you just kind of ignore.

Take for example a fundamental usecase: working with a blob storage service through a client library.

Googling "gcp cloud storage java example" and "s3 java example" returns these as top results:

https://cloud.google.com/java/getting-started/using-cloud-st...

https://docs.aws.amazon.com/sdk-for-java/v1/developer-guide/...

The AWS doc is clear and gives me exactly what I want: Java code snippets for doing common actions with buckets.

The GCP doc is about a sample bookshelf application? And it starts off by creating a bucket through a terminal command followed by a couple of massive code snippets specific to the example app?

Changing the search terms a bit reveals this:

https://cloud.google.com/storage/docs/creating-buckets#stora...

That's a bit better, but seeing options for the various arguments requires me to copy and paste a non-clickable link from the code snippet, which brings me to another attractive but minimally helpful page.

Another example: renaming a bucket. I find the link easily enough..

https://cloud.google.com/storage/docs/moving-buckets#storage...

But only the first code snippet has code samples of any kind. The remaining steps only have instructions for the console and terminal (and the JSON API if you want to curl against that..)

To me, GCP's documentation just doesn't surface information in a way that's conducive to getting work done quickly.

AWS is closer to "copy paste this big blob and modify it a bit". A lot less work for people trying to implement. I'm not planning to read all the documentation, I just want something that works. Google almost forces you to in many cases

If I’m looking at a doc I’m here because I want to do a specific task. The documentation should be task centric like the instructions for a university homework assignment. Imagine you want 30 people to practice a novel task in one evening. You taught them theory in class. Now you want them to try it out. Give them step by step instructions they can use to apply general knowledge to this specific case.

When I’m reading about load balancers and certs I don’t need a textbook. I need examples of specific commands. Instructions on using your tool to do my specific task.

In Google's case, I see a few major issues constantly with their documentation.

All of their documentation is focused around "use-cases", so to speak. Finding complete API documentation is a pain in the butt for many of their services; instead, they present sections like "Here's how to do X" "Here's how to do Y". More often than not, these pages are structured like help wizards, beginning with a "pull this example repository", which is the laziest possible way to check a "write documentation" checkbox.

Barring pulling those example repositories, they often put code snippets in-line with the documentation. As far as I can tell (at least for Go), these code snippets are auto-generated from comment directives in the source code which enclose blocks to paste into the documentation. What this means is that its ALWAYS missing crucial things, like import paths. So, its basically useless; you need to look at the example code to do anything.

Overall, I'd take AWS's documentation any day, though it does have some very bad areas (specifically once you start treading out of the SDK and into service libraries, like aws-xray-sdk). No one is perfect, but AWS is the gold standard of any cloud service provider documentation I've used IMO.

Amazon costs go down routinely. Google costs go up or the services go away.

Specifically, googles inability to adopt a customer-service oriented mindset and customer-privacy will be its end. Is search really it’s core strength anymore? I’d think it’s more it’s ability to reindex in a day and filter out spam.

Fundamentally, as long as google offers free services funded by converting its user data into anonymous normal distributions to be sold, I don’t think google will ever be able to overcome its stigma as a non privacy focused company. I don’t care how many privacy menu settings or SPA control panels it offers. It’s like asking if fb is ever going to be trustworthy.. haha

When a new company scales out a better search algorithm updates as fast as google... on top of aws infrastructure.. that would be interesting

I wonder what is amazons weakness that will be its undoing

It’s been ten years.. still waiting for google to give me a reliable customer service phone number on the quality of amazons.. heck I’m still waiting for google to offer a user service where I know it won’t just die or change 180 arbitrarily. I think it’s a little too late by now. Does the general public even trust in google being a secure, reliable company? Doesn’t seem that way imho

I thought the core business was vacuuming up personal data for ad-targeting resale purposes?

I would love for them to actually a nicer tech company that offers services like GCP and leaves the rest of us alone. Cynically, one wonders how the idea made it past a whiteboard in a meeting, where probably someone was very interested in how they could do some ML on the data coming in and out. Not saying they do (or even could) do this, only that this seems the core business model.

Microsoft was one of the first major developers for the Mac when it came out originally.

Microsoft is happy for you to run Linux in the Azure cloud. Microsoft added Linux emulation to Windows. Microsoft gave up on Windows Phone (if only because Verizon and AT&T refused to approve new Windows Phones on their networks.)

A story that Microsoft honchos loved to tell was that some VIP dared to meet with Steve Ballmer and showed him something on his iPhone. I think the moral of the story is that Microsoft had changed at the time because Ballmer was insulted, but didn’t toss a chair at the guy.

The not invented here syndrome at that company is/was amazing. Microsoft wasn’t as bad as IBM, but they were on that path!

If anything Golang is a counter-argument to your thesis about why GCP isn't taking off. You're right that Google has done abysmally with its commitment to keeping services running for the long haul. But Go itself has been remarkably stable and compatible over the years since its introduction. Far moreso than Python, Rust, or JS in the same time period.

https://github.com/joewalnes/go-getter

IBM Cloud is a slowly sinking ship with the only brightspot being Watson, but nobody seems to be able to describe that clearly so there's that.

I’ve been thinking about building a 10yr hosting plan for my own want. I thought designing static and then hosting on RasPi behind CDN would be great, just stock enough spares and keep Internet service live... 3G or better sufficient.

This doesn't necessarily detract from your point, but this hasn't been true since the release of module support in go 1.11 last year.