Edit: I'm getting downvoted, but no one as of yet has shown how Azure is built on horrible tech compared to its peers. It's just a flat lie, none of the big 3 are built on horrible tech. It's just being dishonest and is basically fanboyism/hatred of one party. You can argue about UI you like or don't (I prefer Azure) you can argue about APIs, their IaaS and SaaS, you can argue that Azure isn't as good on the "edge" compared to GCS, you can say kubernetes is a bumpier ride on Azure etc, but saying the tech is "horrible" is not correct. Sorry. This isn't early 2000s Slashdot.
I don't have familiarity with Google Cloud though. And in fairness I used Azure for a feature AWS didn't have anything close to (service bus queues), I can see why having really well built things like that get them huge deals. Microsoft know what they're doing in those markets.
GCP console and products are very consistent and well planned, but the console itself is slow to load because of their heavy UI framework and dumb animations. I much prefer ugly and fast over clean and slow.
AWS is messier than Azure but faster than GCP.
I still don't know how Google can be so bad at creating responsive user interfaces.
1) Disk and network IO performance was extremely inconsistent, to the point of being unusable at times. 10kbps reads off of the msft-local mirrors of package repositories for example, making security updates a tedious pain in the ass.
2) There were two administrative consoles, "new" and "old." New had a more modern-looking UI but didn't reliably work, old was cruddy looking but did. The set of features present in them was not congruent either ("new" had some some "old" didn't and vice versa). But wait, there's more! see #3
3) Not every operation was possible via the admin consoles. Some things you had to use their powershell cli tools to do, which is great fun when you have no windows machines around to run powershell on. This was for something stupidly obvious like "assign an IP to this instance" or something; mercifully I've forgotten the details but it was something you'd think would be trivial.
4) Whoever designed their payment model was ... to be charitable, extremely set in their ways. Instead of being able to set up a payment method and pay by the hour (or whatever), you had to buy "entitlement packs" at $X per license and apply those to your account, sort of like the boxware model but awkwardly shoehorned into cloud billing. Woe betide you if your "microsoft bucks" ran out in the middle of an extended compute run!
All these were for comparatively simple and straightforward uses of their infra (i.e. pretty much just compute and block storage, no fancy database or machinelearningAIwhizbang-as-a-service stuff). I can't imagine the more complex features were better off if the foundations were so haphazardly implemented.
2) This is no longer a thing, though during that transition period it was slightly annoying.
3) This is also no longer a problem.
4) I've never experienced this, the "pay as you go" model has been with me since I started using Azure 6-7 years ago. Maybe they do this for some licensing models? I have only ever paid for compute time, storage usage and egress. I've have used plenty of licensed software too...so I don't know what yours relates to.
Back to #1 though, if the Azure team could figure out why Web App disk speed feels so slow (sometimes), my company would probably double/triple our spend on Azure and drop some of their competitors. This is genuinely my biggest complaint.
Speaking as an SRE who less than 1yr ago was tasked with evaluating public clouds as burstable capacity for our traditionally bare metal server infrastructure; this comment is flippant at best and harmful at worst.
Azure is absolutely not comparable to the others, their performance characteristics are nowhere near consistent between equivalent specification instances, their API's are equally inconsistent _and_ they have a terrible usability model on most of their services (not all, admittedly).
In fact when it comes to technical competence, I would (and did) rank Google #1. The drawbacks of Google are:
* It's google and they have a habit of sun-setting products.
* They don't have as many features as AWS.
* They don't have developer mindshare like AWS, meaning FOSS tools will almost always work flawlessly with AWS but rarely have support for GCP (or, if they do it is a little b0rk)
* Google tends not to give human support. (but this is alleviated if you're buying support contracts)
-
FWIW we chose google on technical merits alone, although my company is working with all three cloud providers in some fashion. Azure is the one we constantly mock internally for their absolutely maddening warts. Almost as bad as our own internal "cloud". (providing cloud services is not my companies core competence to be fair)
----
Digression;
I would assume that a big chunk of Microsofts cloud money is coming from office365. I know my company recently started paying them in the order of 10's of millions of dollars, I assume others would too as this is "the future" of microsoft exchange/sharepoint etc;
Honestly, it sounds like you just don't like Microsoft more than an even keel observer.
My bias is purely on the technical merits of the provider. My company had a pretty large discount on Azure so it was under strong consideration.
My point about it "being harmful at worst" is that it's spreading uncertainty without any actual evidence.
I, on the other-hand have evidence from a 12,000+ person company that is using all three cloud providers.
Actually, that's precisely what -you- are doing, not me.
> I, on the other-hand have evidence from a 12,000+ person company that is using all three cloud providers.
I worked at a much larger company than that with over 20K employees that used Azure/O365 and AWS (but no GCS.) Your anecdote means nothing. You're on Hacker News, there are people here that work and have worked for massive media companies, industrial companies and tech companies. Many of which use Azure, AWS and GCS.
Do I wish Azure were better? Absolutely. I have a lot of ideas and complaints where things could be so much better. But I also have just as many numerous complaints about GCS and AWS.
FWIW our migration to office365 has been pretty great, but I don’t see it as being the same thing as using Azure for your product.
No it is not! Azure is not even in the same ballpark. APIs fail randomly, instances take a random amount of time to come up(sometimes similar to AWS, sometimes double digits) services are poorly integrated, lots of weird constraints and surprising behavior(their load balancers are nonsensical), lack of AZs (and support for them) in many regions, etc.
Can you make your stuff work on Azure? Sure you can. Is that a good experience? No way.
This is probably what people are referring to. The underlying tech may not be horrible, but the user experience is.
First of all, I'd want to acknowledge that Google is behind AWS as well. One big technical reason is IAM. Few realize just how important AWS IAM is as a service federation infrastructure. If you look at the details, Google's IAM product is inferior.
Now, on to Azure. Three years ago I was involved with a high stakes effort to port an AWS-grown platform service architecture onto Azure. At the time, they had massive gaps in their understanding of what IaaS meant. Here are some concrete examples:
* They did not really understand what object storage was. Blob storage was not possible to use at scale due to trivially low bandwidth, storage, and API limits
* Software-defined networking was not available between availability zones
* Software-defined networking could not be used to launch mixes of instance types
* Software-defined Internet gateways were not available except in a config that resembled "AWS Classic" networking
* On-demand instances were effectively unavailable beyond one or two instances at a time (at least for the instance types we wanted). You had to reserve instance capacity in advance, by going through a support ticket
* Creating and using custom machine images was undocumented in the API
* Instance metadata APIs were not available
* On-demand instance launches would encounter weird behaviors, where upon hitting certain limits entire groups of instances would be terminated
* Many aspects of APIs for the above were undocumented and unsupported
Combined, these problems made deployment on Azure extremely difficult. I have prefaced this with the caveat that Microsoft has improved since then. Many of the problems above are no longer issues, I'm sure. But what I found was a gaping chasm between what Microsoft claimed and what was really possible on the ground. What I found since then is that Google and Microsoft are making an earnest effort to catch up, and that's good for us consumers, but Microsoft (and to some extent Google) often don't even understand the full feature set of what they are trying to catch up with.
Things change fast in these cloud services - so would be good to know specifically which parts of GCP IAM are missing.
Obviously, my experience is rather shallow here and wouldn't dare go into too deep discussions on that. And the shoving part can easily be avoided with some extra elbow grease.
Ideally cloud compute remains a commodity resource and all players earn a standard and pretty average return on their capex..
My anecdote is to Warren Buffet - buy commodities and sell a branded product. Don't get caught in a trap where you build your product with unique cloud resources which you can not easily migrate to other platforms when the conditions calls for it (price, performance, or even support.) Right now, there may be endless VC dollars, but there are some markets where the long term winners are going to win because they are paying commodity pricing on all of their compute.
Further you can use Powershell (from the UI, no less,) if you are all about CLI.
But every time I deploy a new AKS services, it takes 30 min for it to be ready. I luckily haven't had to do this recently (~4 months?) and I'm loathe to try again. GKE on the other hand is within 2 minutes.
Provisioning instances take quite a long time. The response I received from a product manager was, "how long should it take?"
Regarding MS taking 2nd place - recently came in contact with their cloud. It's not up to par to the rest from a tech perspective, however, they are killing it on their sales channels and in sectors such as banks and retail due to their ubiquity in those.
Edit: Wording, horrible is not the best description of the MS cloud, just not as good for me.