103 comments

[ 2.6 ms ] story [ 178 ms ] thread
"Anthos clusters on AWS" No equivalent alternative.

Uh. Really, no Anthos on AWS equivalent on Azure? Or AWS? Boy, they're really missing that managed Kubernetes cluster game.

Maybe I'm just dense, having only been doing this cloud thing for a decade or so, but this table seems...misleading or just wrong in so many ways.

Keeping scrolling! Under GKE (managed Kubernetes), you see the call outs to EKS and AKS, respectively.

The Anthos point is that other clouds don't offer their k8s stacks as supported options on other clouds.

If you see things you find wrong in the table, let me know.

Disclaimer: I work in product management at Google Cloud

Does anyone actually use Azure Arc?

Azure Policy to me seems like a tech demo more than a polished product. Even for common tasks such as enabling Diagnostics they just provide some "sample" policies, and then authoring the other 275 distinct policies for each resource type that can be monitored is your problem.

Does anyone have any real-world experience with it?

Arc looks cool, but it's only the control plane. They aren't dropping AKS into other environments. The compute is your responsibility. Anthos puts GKE in other places.
Not sure if it's already GA, but AWS announced a product called EKS / ECS Anywhere last year that covered this use case.

Edit: not GA yet

CloudBuild, CloudDeploy, CloudPipelines

guess it should be CodeBuild, CodeDeploy, CodePipeline

> "supported options"

So you can get Google managed k8s on azure with Google support (ie: none?) rather than managed k8s on azure with ms support (ie: some?)?

(comment deleted)
Is there a thing where I describe my workload and it will give me price ranges for all the clouds?
I don't think Google's fabled AI could even work out costs in any cloud service.
Google can't even work out costs for their own products. For example try asking the pricing calculator to quote for a shared core Cloud SQL postgres instance: there's a drop down that invites you to choose between shared and dedicated types but neither is an option, only 'custom'.

https://cloud.google.com/products/calculator

Taloflow got featured here last week - might be what you're looking for? https://www.taloflow.ai/
Looks like it's in the right direction, though I don't like "write speed: 4 stars". Seems like it just be in Gb/s rather than subjective star ratings. Thanks for the link.
The resource limits are so severely restricted on Cloud Run that I don’t think it’s fair to compare it to Fargate. The space of use cases solved well with Fargate is far, far larger.
> The resource limits are so severely restricted on Cloud Run [...]

Honest question: which of the limits in https://cloud.google.com/run/quotas#cloud_run_limits do you find “severely restricted”?

Maximum memory per instance (8GB) is an extreme limit. Disk and CPU limits per container instance are also quite bad.

And, laughably, for any workload just a bit out of reach for Cloud Run, GCP docs immediately recommend switching to GKE (and even Anthos).

Imagine having a high RAM workload that is just a simple RPC service. Many (probably most) machine learning services fit this model. Many routine ML models require more than 8GB RAM just to load the model, bit it’s a good use case for serverless non-lambda infra because it runs out of a very unique Docker image and does nothing but serve stateless model predictions.

Needing to bring in all the machinery of GKE or pay out the nose for Anthos just because you need exactly the same operational model as Cloud Run just with high RAM or CPU is a really poor customer experience that feels deliberately set up to push you towards more expensive Kubernetes products.

This list is biased, its missing Azure products like Azure Data Explorer, of which AWS/GCP has no alternative (afaik).
> This table lists generally available Google Cloud services and maps them to similar offerings in Amazon Web Services (AWS) and Microsoft Azure.
The title of this post implies otherwise: Compare AWS and Azure Services to Google Cloud
Can you not use Bigquery or Aws Redshift? On the microsoft blog posts it seems Azure Data Explorer is more price efficient, though I don't quite understand besides that is it doing something special compared to microsoft's existing offerings?
ADX is an extremely good timeseries data store. Integrates very easily without hassles into EventHub , Blobstore etc, and also has a very intuitive query language. (Though we could say AWS product integrate with Kinesis / S3 into RedShift). Needless to say a lot of nice features like Geo queries , extremely good support make it really good!

Azure was not as good as GCP in terms of data offerings , but with ADX and CosmosDB etc is really getting par with other clouds really well.

Coupled with strong Azure IoT , these data products may be a real winner

Yeah but...then I’d have to use azure.

The last time I was subjected to azure, I wanted to drive a stake through my head out of sheer frustration.

Curious to know about your experience with Cosmos though, when I last looked it was eye wateringly expensive and reports were that performance was...slow at best.

It is PaaS , you are locked down (just like a lot of other solutions.) The biggest problem we had with other no-SQL databases (we were high on interop and started with HBase / Phoenix ) is that over time the storage costs and operational costs get out of hand especially if we do not have a good data archival part (sometimes limited by legality as well). As for your frustrating experience , the mileage varies. We were happy with it , so no comments on that one. For cosmos , we used it for extremely "hot"/"warm" data that was aggregated and made available readily for applications to use. We had reserved instances (big on Azure) , so we keep things in check that way.
Competition is always better and welcomed.

But I really don't understand what edge google offers other products. If anything, I am always really wary about google and their products. People might say that enterprise services are different, and it's different from free services where if you're not paying you can't expect anything, but there's been paid services by google that have been absolutely terrible like stadia. There's something wrong with Google as a company that I can't think of much that they've done recently that seems solid. There's a reason why managers get paid high salaries, because it starts at the top and flows down.

At the very least the marketing for gcp could make it clearer what exactly it's offering that's better: price, ease of management, etc. Because nobody is going to get fired for choosing AWS so there has to be a real clear incentive other than some small technical improvements that don't matter at scale.

Stadia is a huge project, backed by google. I think it has “canary in the coal mine” features. It does want to be an enterprise service, eventually.
We'll see how it goes with Stadia. I think it's a fairly terrible product. Google doesn't keep things like it alive for very long.
Stadia is the best streaming game service there is. People just don't want that product.
It’s not significantly better than something like rain way, which at least has the courtesy to plug in to your pre-existing games library.
And, if i'm reading their site correctly, requires you to have a gaming PC with good networking. It's not even in the same ballpark as Stadia!

What you should be comparing Stadia to is GeForce Now, Amazon Luna, Shadow, which provide the computing power as a service. Stadia eschews Windows to run the games, making it cheaper ( their free tier gives you 1080p/60fps for free, you just have to buy the games on the Stadia Store) but limits what games they can run ( they have to be made for Vulkan). The downside is that you have to buy the games for it, but it supports ubisoft+ ( with an addon on the Ubisoft side). But you don't complain when you have to buy games anew when you buy a new platform ( be it a PS, Xbox or PC), and the same logic applies here.

GeForce Now and Amazon Luna are still a Windows server box behind the scenes, with the associated slowness and a bit clunky integrations ( e.g. GeForce Now requires you to re-login to Uplay or Epic or Steam every so often, which is painful with a random password and MFA, and no copy-paste available).

Shadow provide you full access to a Windows box, with the associated issues ( game updates and downloads are up to you, disk space is limited, etc.).

Stadia has some downsides, but IMHO fares very decently against the competition. I have both it and GeForce Now, for different games, and IMHO Stadia's UX is slightly better.

I love Google’s Cloud services. I think in many ways they have the best product.

I also worked somewhere where we made a million dollar commit. We had to get support through a VAR because Google basically doesn’t offer support. We basically paid someone else to file tickets that got dismissive responses while we had prod meltdown.

They’re also price competitive. I was going to go there again in my current position in a very small org. We have GSuite and to get GCloud working we had to turn on Cloud Identity. We didn’t see the free tier in GSuite, so I went to ask GSuite support. The only accessible support was a bot. When I asked the bot how to talk to a human, it helpfully referred me to a bunch of forum posts of people complaining you can’t talk to a human at Google.

We’re now with Azure.

I'm looking for a managed Knative solution and Google Cloud Run looks like a great product. The only reason I'm not using it is because I also had first hand experience with Google's atrocious customer service in the past.
In what ways is Google the best product if you had a prod meltdown and couldn't get support, and when you wanted to increase your use of GCS at another job, you were unable to do so?
> In what ways is Google the best product

> I think in many ways they have the best product.

They didn't say that Google had the best product. They said that in many categories, it was the best. That doesn't mean it's the overall best product. In fact, from their post it seems pretty clear that they ended up deciding that Google Cloud wasn't the overall best product.

> if you had a prod meltdown and couldn't get support

Presumably, they were implying that customer support _wasn't_ one of those many ways.

I think they were just asking them to justify in what ways do they think Google have the best product as their post only really highlighted the ways they weren't.

Given the experience they had in the ways they were bad, it could be difficult for others to justify them despite the ways in which they might have the best product.

I read it as completely steeped in sarcasm
Yeah fair question.

I love their GUI, the way projects are organized and RBAC is handled across them. Creating custom cpu/mem ratio machines.

Little things like VMs in a kubernetes cluster being viewable as VMs (they’re obscured in Azure I’ve learned).

Pricing!

Like, there are very few ways they’re not my favorite cloud provider. Which speaks to how unspeakably horrible their support is that at my current org I am not going with them.

Because at the end of the day, a pretty good infrastructure product that you can get support for wins every time over an excellent one that when it breaks is catastrophic and nobody listens.

I’m spending this week rebuilding all my Azure infra because I learned about 10 different ways they hide ridiculous price increases in what you build and my infra costs twice what I thought it would. But I literally can’t get ahold of someone at Google to help me become a customer.

Meanwhile I’m not even paying for support at Azure yet and when I sent an email about the pricing I got a response from a human who checked up on my satisfaction, and then from her boss, who was checking on my satisfaction with the first email.

How long ago was this? On my ‘basic’ workspace plan the help button asks you to describe your issue, but then it’s one click away from getting phone support.
Hey, now. At least you don't have to dismiss the chromebook notification from the old support portal...
The phone and chat buttons never show up for me. We’re on a non-profit plan so maybe that’s why.
We've used Google Appengine for over ten years now. It was the first service to abstract infrastructure, so developers could simply deploy code without worrying about scaling and redundancy.

Google's edge (for us) is reliability. Over our ten-year tenure (heh), there have been a handful of incidents when our service was down, most early on before the NDB datastore. I seriously can't remember how many years it's been since our service was down.

And when Appengine does break, they do a brutally honest report about exactly what happened, and how they are going to make sure it never happens again. If you've got the interest and time, it's worth searching for articles/videos about Google's SREs. They are fanatics.

I know this is anecdotal, but AWS and Azure seem to have multiple outages a year. Every time I'm immensely relieved to be on Appengine.

Google Cloud is just a pleasure to use. It has by far the best documentation, the best user interface, the best CLIs (gcloud etc) and tooling, and they have the most accessible teams of any of the other cloud providers. I remember having an issue with firebase authentication a while back, I posted a question on stackoverflow and a firebase team member had answered it within a few hours. Good luck getting any help with AWS Amplify when you encounter any bugs/issues.
Hm AWS support has been very good to me. Granted we pay for it, but every time I had a question they actually reproduced the situation based on my descriptions, prototyped a solution and came back with exact steps on what to do, pretty quickly, too.

(Can't speak for GCP or Azure, I run limited production stuff there since historically we just happened to get a ton of AWS credits first. But, I try to stick with open-source instead of Amazon's proprietary just so that migration or multi-cloud remains an option for us)

It speaks volumes that you thought to get your gcp support from stackoverflow. Good that it worked that time, but plenty of SO questions go unanswered, they aren’t private, and there’s not much in the way of guarantees if you have a big problem on your hands. I don’t always ask for gcp support but when I do I’m on my own. Like that one time they shut down production because they didn’t like our credit card card anymore, because of a mismatch in nationality and the UI not offering any way I could change the nationality of our account (we were paying for it with my Dutch credit card, but the company is registered in Germany—our German credit card is prepaid, and GCP did not allow such cards), and obviously getting a new credit card took time. I was very relieved that this wasn’t for our main service (on aws), but the statuspage for it, hosted off-aws for obvious reasons. Now, I don’t always ask aws a question, but when I do they are able to timely help, whether with a technical or billing issue.
> I am always really wary about google and their products. People might say that enterprise services are different, and it's different from free services where if you're not paying you can't expect anything, but there's been paid services by google that have been absolutely terrible like stadia

GCP is a major player in the cloud space. Number 3 last I looked. I fail to see how that compares to Stadia, an expensive experiment that may or may not succeed. Do we judge AWS on Amazon Prime Video's mediocre service? I sure hope not. Let's be clear - GPC is not an experiment. It has massive capital backing

> At the very least the marketing for gcp could make it clearer what exactly it's offering that's better: price, ease of management, etc.

Okay, but you (I assume) can read beyond marketing. Marketing should not influence your decision (very much)

> Because nobody is going to get fired for choosing AWS so there has to be a real clear incentive other than some small technical improvements that don't matter at scale.

I don't use GCP but there are many clear advantages. If those do not help your biz case, fine, move on. But if you can't see the advantages, read more

You're talking past each other with the word "marketing". One of you means "messaging and product positioning detail" and the other means "biz-speak drivel"
>But I really don't understand what edge google offers other products

Google Bigquery is head and shoulders above the competition, a really impressive work of engineering. It moreover sidesteps Google's poor API design culture by using SQL instead of something home-grown.

My personal experience with Azure/AWS is that sometimes the products sound the same but there are painful caveats.

For example, Azure Event Hubs and AWS Kinesis both say you can pick a shard key (eg CustomerId) in your messages to have processing order for each customer.

However, Azure Event Hubs notes (and my experience was that) using this feature is subject to downtime because each shard is not highly available so you can't really use it as the first step, but it is reliable on Amazon.

It's true, no two products are ever exactly equivalent. And also you might like a specific product, but avoid using it because you don't like the platform it is part of.
After installing the GCP CLI tool, I keep having this nagging concern that it would be used to track me for advertising purposes.
There's a configuration option to disable that, as per Google's documentation: https://cloud.google.com/sdk/usage-statistics
Should be more opt in rather than opt out. I know some Azure libraries do something similar , but with google , i do not want advertisements on youtube about "demand of AI engineers and self driving cars" because i use gcloud CLI :)
> Unless you opt-in during Google Cloud SDK installation, Cloud SDK software does not collect usage statistics

It is opt in.

(comment deleted)
(comment deleted)
Every time I am tempted to use Google Cloud, I wander over to https://killedbygoogle.com/ and regain my senses.

I am not saying that Google is going to kill core product offerings like K8S... what I am saying is that they lack focus, and change direction without regard for their customers. That, coupled with their famously bad support means that I can't sleep at night with the confidence that I can build a complex product on top of their offering. I'll always be worrying that I have to rip out or re-architect large parts of my code to deal with their lack of product stability.

Oh just get your Google account shut down with zero warning and loose everything forever.
This is and the lack of human support is the reason I w ill never host anything with Google ever.
While I understand the sentiment there is actually human support available on Google Cloud Platform, and there are contracts and SLAs for service availability and support.

Also, as part of Google Cloud for Startups we were assigned a local (as in: operating in our country) account manager that periodically checks in to see how we're doing and connects us to Google's partners where applicable.

We're actually very happy with Google Cloud Platform, and have been for the past ~3 years.

For me it would still be questionable how quickly they react when some of their algorithms wrongly classify one of the applications I host as malicious.

And given how easy it is to e.g. trick the Youtube algorithm to believe a video is hot using fake engagement, I very much dislike the worldview they have that algorithms can solve everything in 2021.

Not sure what you mean about a lack of human support, but i've had a great experience with their support team so far if it was ever needed.
About 8 years ago I migrated the company's infrastructure to AWS. About 5 years ago I built out a real-time Android app and decided it would make total sense to use Google Cloud services. I regret that move even today. GCS services are unreliable w/many limitations, let's not even talk about UX/UI. AWS is not all peaches and cream either, they launch new services all the time (which are not so well developed ie. IOT / pubSub /) but their core services are solid (S3/CF/SQS/SNS)
When I first started dabbling into trying to do something on my own, Google App Engine (GAE) seemed the easiest to understand and get off the ground. And the fact that everything 'just worked' and didn't require setting up stuff was awesome in my opinion. GAE Launcher was the cherry on the cake which is why it was such a bummer when they deprecated it (and this led to my current side project - https://nocommandline.com - which is a replacement for GAE Launcher).

I was initially worried GAE wouldn't survive but finding out that Khan Academy is hosted on GAE [1] gave me confidence of its longevity.

I once tried looking at AWS and it took me awhile to figure out they had a PaaS and trying to set it up essentially gave me a headache (keep in mind that I'm not a 9-5 programmer).

1. https://cloud.google.com/customers/khan-academy

It astounds me to this day that AWS doesn't have a heroku-style "five clicks to have a github repo deploying at a url" PaaS offerring.

Beanstalk ain't it, and they cloned DO's "create a VM wizard" (Lightsail) but still no simple GitHub oauth webhook build/deploy.

PaaS is such a great and convenient model. It seems a major oversight to me on AWS's part.

Doesn't heroku run on aws? Ie: aws makes money from heroku?
Isn’t Snapchat also running on GAE ?
What would be the motivation to publish such a page on the Google side ?
They are encouraging switching?
But they still advertise aws and azure a bit right. Isnt there a risk of driving google customers to aws or azure ?
Google's challenge right now is to get customers from AWS and Azure. People who will see that page are probably comfortable with AWS or Azure and so Google wants to tell those people that we have most of the features which our competitors have.
Makes sense thanks.
Why should I trust Google to compare GCP, AWS, Azure? If I really want to compare services, I would do it on my own, look for independent websites or pay an independent company to does it for me.
To be fair this is not a comparison, but rather showing equivalent products with no qualitative rating of the other companies’ product. Therefore I think this is a good Ressource albeit not a comparison.
In a lot of rows they claim "No equivalent" available. If a company tells me that there is no competitor equivalent for one of their products, I would say thank you and resesrch on my own.
I use GCP for hobby projects because it’s simpler than AWS, has a better GUI and better documentation.

For a commercial project I would probably go with AWS, since I’ve heard a lot of horror stories about GCP support.

I’ve been scared away from Azure after using it for a couple of courses at university. It just seemed half-baked: lots of GUI glitches/bugs (e.g. reliably needing to press a button exactly twice). It just seemed like one of those Microsoft products that only exist because the best product in the category isn’t owned by Microsoft already.

I've had good experience with Linode for my personal projects. They have a lot of cloud products nowadays like S3 buckets, so its no longer just VPSs.
Azure by far has the best customer support, great for commercial use
Ive used Azure and AWS in a couple of large organizations (mulit-million $ cloud contracts) and the Azure commitment for customer success is a big differentiator. Ive literally had Azure engineers as an extension of my team (free of cost offered by azure as part of customer success teams) so much so that they start blending into the team. The Azure products have steadily gotten better (ADB, Azure SQL, Functions etc) over the years (with a few duds as expected). AWS - it didnt matter how big we were, we still got the feeling they never really leant in or they had "more important" stuff to focus on than customer success. All the POCs/side projects we have done on GCP - it has always felt like something built by a bunch of seriously talented technologists FOR technologists. Support seemed a distant entity just like most google products - that has led to significant nervousness at the top leadership to invest in a product that may not be Google's highest priority.
GCP has vastly worse documentation than AWS unfortunately.
I have to disagree with that, I've found GCP docs to be much more informative and easier to digest.
I find the AWS docs too verbose, however it’s quite possible that doing something in AWS is just more complicated than doing it in GCP. In that case the issue is not with the docs, but with the complexity of AWS vs GCP. Either way GCP has an advantage here.
>It just seemed like one of those Microsoft products that only exist because the best product in the category isn’t owned by Microsoft already.

They've bet the future of the company on it, it's definitely not a product that only exists because the best product in the category isn't owned by them.

Short of Microsoft as a company going under, Azure won't be going anywhere. There are countless companies that will not do business with Amazon due to them being direct competitors and Azure tends to be the place they land. As others have noted, while there may be some technical gaps between Azure and AWS, they more than make up for it with their legacy enterprise domain knowledge and support (in my experience).

GCP under Kurian appears to be trying to cater to that same subset of customers, but the culture of Google still seems to be a pretty major uphill battle when it comes to being "enterprise friendly".

isn't azure the only cloud provider that hasn't censored any of its clients, by disconnecting their services, yet?
Personally my experience with the Google Cloud has been the best, but I think if one goes with AWS or GCP they'll be fine either way.

Just stay away from Azure. If death by a thousand needles was ever true then it must be for Azure. The hidden caveats, pain points, down times, subtle bugs, Portal issues, confusing documentation pages, and so on are a HUUUUUGE time sink. Honestly the cost of Azure is irrelevant when you spend 5x more expensive developer time in getting simple stuff working.

Don't say I haven't warned you...

Azure has absolutely the worst interface in my opinion, but I’ve found that for a lot of my clients them already having an azure account (via AD) outweighs pretty much anything else.

I think Microsoft know this and know they come a long way without having to compete.

Honestly, I liked Azure's web UI better than AWS's.

It felt easier to navigate from one service to another.

All 3 clouds have their warts and the grass will always seem greener once you pick one. Hard to go against AWS based on size and head start though...
There's a lot of truth in this. For large organizations though having a multi-cloud strategy from the start is a smart thing to do. We got hit pretty hard from the AWS outages last fall, so our previous 'all in on AWS' approach has been replaced with 'mostly AWS but let's look at GCP and Azure'. We spend millions per month in AWS and the incident last fall broke trust.
I'd like to see columns for "avg time to talk to a human(support)" and "recourse on account deletion due to perceived policy violation".

I'm not sure how azure and aws score - but better than Google?

[ed: as a side note - I've yet to try and get support from azure or aws (we use both) - but we did get prompt reply/action from digital ocean the few times we've needed to reach out]

I really would not consider Dataproc to provide the equivalent capability of AWS Batch. It's quite painful how many of GCP's services rely on GKE. Ultimately I'd like to see more services like Cloud Run from GCP.