Ask HN: Which cloud do you use, and why? | Hacker News Reader

Ask HN: Which cloud do you use, and why?

53 pointsNelkins83 comments

Hey all,

The recent HN post on the "$10 super computer" (https://news.ycombinator.com/item?id=9575683) was actually very timely, as I am working at a client in the financial space who is evaluating various IaaS providers (up until now they have been using a small on-site data center). There is plenty information out there about which services are available, but less about why one should choose any particular provider over another. I'm curious to hear about the experiences of the HN community on this, and especially from those who have used several different services. What sealed the deal for you? Was it the price? Ecosystem of services? Ease of use? Was it the only thing available at the time? etc. Would also love to hear about any other factors one should take into account when making this kind of decision.

View on HN

AWS currently and for the last ~4 years. We have also placed more than a dozen clients into AWS.

We have used SoftLayer, Peak10, Google and Azure at different times while testing things out. SoftLayer we actually ran some processes in for about a year.

I recently had posted on HN about services like DigitalOcean because there are ways for us to save money using those types of services, or even Heroku. In the end the ecosystem, ease of use, reliability (when architected right) and flexibility of AWS is still just too powerful for us. The costs are not the cheapest way we could do something, but when you factor in the minimal amount of man hours we spend on infrastructure compared to what we could be, I think it is a good deal. We try not to get too ingrained to AWS only features, and where we do we will hide their implementation so we can switch them out if ever needed. I also classify them differently too, for example, DynamoDB or MySQL (via RDS) wouldn't be something I'd try to wrap, it is your database and could be used from anywhere so you aren't really locked into AWS as much as your database there. But something like SQS, we used but wrapped, and eventually moved to RabbitMQ with very little extra effort.

I have had to use AWS support a few times, and they have been AMAZINGLY helpful. And for non-critical things they usually answer forum posts pretty quick too. AWS isn't a panacea with no issues, but for my money its the best thing going right now. And they are constantly releasing new things and tweaking things, not everything is for the better, but for the most part it is.

Heroku for backend, automatically grow or shrink resources, nothing to maintain, nothing to configure and tons of services available alongside it.

Netlify for frontend, it is a static hosting service with some neat features like url rewriting that lets you proxy e.g. /api to a backend without revealing api keys and under the same origin / domain.

What really draws me to this is:

- no servers or software to maintain

- horizontal scaling by design (heroku kind of forces it)

- deploy process is "git push", fresh deploy is one click (heroku) and a few mins doing config on netlify though they have an api so maybe that's redundant

- no lock-in, easy to replicate locally

- databases are managed wrt backups, recovery etc

- entire front end is static and will scale indefinitely with massive uptime

https://heroku.com/ https://netlify.com/

Why netlify and not S3 + CloudFront?

The final outcome is the same - static files on a cdn - so it won't always matter but netlify lets you:

1) run a build process

2) own domain with automatic ssl and redirecting to w/e www-or-not, https-only etc

3) deploy with git push

4) url rewriting, header inserting, proxying

Exactly, and because of the automatic URL rewriting Netlify will also give you instant cache validation. So when you update your site, unlike most static site hosts, you won't have to wait to see your changes live. But at the same time you get a very cached and very fast site. So best of both worlds.

We have an article on our blog about some of the differences.

Typically you'll have to do quite a bit of work to get the same performance on S3+CloudFront as out of the box with Netlify, and if you do set up things to be as fast there, you'll suffer from slower cache invalidation.

https://www.netlify.com/blog/2015/03/06/comparing-netlify-an...

What kind of traffic does your heroku instance handle? How many dynos and which dynos do you use?

Azure, awesome PaaS services, very easy to set up, good documentation, good pricing, very good performance and I really like the fact that they are working really really hard on Azure. Big new stuff is being released every month.

Combining Azure with Cloudflare has really been great for my dozen or so clients I manage.

It ups Azure's performance and adds another layer of excellent security. The Pro plan at $20/month is well worth it.

I'm using Azure.

Mostly because of this very little known offering called the Microsoft Action Pack subscription: https://mspartner.microsoft.com/en/ca/Pages/Membership/actio...

It costs around $500 per year, offers your company access to a large number of Microsoft software licenses, free access to seats of Office 365 Enterprise (for hosted email on your own domain, in addition to the software), and most importantly, $110 in Azure credits per month.

The Azure credits alone make this an amazing deal that essentially cuts all Azure pricing by more than half, making it more than competitive with AWS/GCE/DO on the pricing front.

I'm a current Azure user so this intrigued me, but I don't see anything on that page about monthly credits. Am I missing something?

Edit: Never mind, I guess? I found it in some fine print on another page. Weird that they don't advertise it more prominently...

Yes it's pretty ridiculous how well they've hidden one of the best perks of the subscription. I only know about it because I was enrolled in a previous version of the program that advertised the Azure credits a lot more prominently.

Their marketing in general is sorely lacking seeing as how few of the people who would benefit from it actually know about it.

For anyone else wondering, I managed to find it mentioned on this page (deeply hidden under "Microsoft Action Pack – cloud and on-premises Internal Use Rights licenses" accordion, under the "Management" section in the dropdown, under the "Cloud Services" heading):

https://mspartner.microsoft.com/en/us/pages/membership/inter...

"Microsoft Azure

US$100 monthly credit. Microsoft Azure credit is in addition to current on-premises internal-use software licenses."

If you are a silver partner as we are you also get a 100$ plan for azure every month!

Digital Ocean

- I only want to build off base Linux VM, I do not want to be locked into S3 + SQS + ...

- Price is right for young startups

- Speed is nice

I love Digital Ocean, I have been using them a while now.

I like the UI, the referral programme and the speed of there whole spinning up process and finally I was once in arrears by 2 months and they were pretty lenient with payment.

Talking about referrals if you want a free $10 - https://www.digitalocean.com/?refcode=c8555bae4f81

DO is really nice. I use a little droplet myself and for the given price of 5-10 USD/month it is a great value.

Once you want something more I wonder if there are better options. Take there 320 USD/month plan. You get 32GB memory, 12 CoreProcessor, 320GB SSD disk and 7TB transfer. If I take a fixed price server at Hetzner I get 128 GB memory, 12 CoreProcessors, 2x240 GB SSD disk and 50TB transfer. Even the smaller machines are much beefier than what DO has to offer.

Once I am running a business 139 EUR {~155 USD} per month should both be fine.

That will always be the case, with AWS if you sign a contract with them it brings the price down compared to on-demand. If your using an on-demand service you will always pay a premium

I +1 the dedicated server.

100£ for such a monster: http://www.soyoustart.com/en/offers/sys-ip-6-s.xml is really reasonable, imho.

Impressive :) thank you for that.

I just wish there was a single VM that wasn't SSD backed. I write web crawlers sometimes and I usually only have to parse the HTML / JS once into JSON or SQL, but I like to keep the data backed up just in case I need to re-parse it later, or to see the changes between crawls. It's completely impossible for me to even fit it on their $640 / month box. But I love Digital Ocean for everything else.

Have you looked into the Kimsufi boxes? Sometimes they are available for 5 EUR/month. They come with a Atom cpu and 500 GB of storage.

That's why I like TransIP. Cheap SSD-backed VM + large network drive (up to 400TB).

Couldn't you potentially push that aspect elsewhere, into a data storage solution like S3 or something (not sure of the cost efficiency, not a huge cloud guy myself). Then you'd just be utilizing the DO server for processing. It would have the added benefit of allowing you to scale your crawling horizontally if you wanted, as all your servers could have access to the same set of data.

The traffic between datacenters would be significantly slower and more costly, though. Besides, S3 is not a replacement for a real drive.

It would be slower than on disk for sure, so if that's a concern, it wouldn't be a good choice. If the poster is just storing off results, pushing it to S3, or even potentially a document database could be a logical choice. They said they're just using it for occasionally re-parsing at a later date, so immediacy doesn't seem like a huge concern. Just a thought, at any rate.

Yeah I could, and I currently do - but I don't like to. I want the data on the same network for lots of reasons and I also don't like maintaining two sets of deployment systems. I'd prefer to just use Cloud66 + Digital Ocean and focus on writing code, not setting up and maintaining servers.

Oh yeah, I totally understand that, was just offering an alternative as a non-ideal workaround. I have the same set up for a service I offer, primarily because (afaik) it's hard to beat S3 pricing.

Vultr, who have offerings very similar to DO, have a class of VMs called "Storage Series"[1]. In my experience their uptime is similar to DO and their performance is better. Might be of interest to you:

[1] https://www.vultr.com/pricing/

Take a look at serverhub.com ssd cached vps, $5 per month for 500GB of diskspace.

I want to love Digital Ocean but is there any talk about providing load balancers? It seems like as soon as you want to scale beyond one server you're stuck rolling your own with something like HA Proxy, but you've still got a single point of failure. I guess you have that with a hardware load balancer as well, but at least those things are specifically built to be available all of the time, versus just a piece of software you're running on yet another VM.

Yeah I totally agree with you on that one, I don't really want to setup another server to have a load balancer. I would rather use a service like how AWS do it.

I'd also like to see a storage server - one with more storage than usual, without the necessary increase in cpu/memory/price.

akinder, it seems your account is "hellbanned" which means nobody can see your posts unless they are logged in and have showdead turned on.

(I'm sorry I have to clutter up the discussion with this, but it's the only way I can communicate this news to akinder)

You should check people's comment history before you tell them about hellbanning.

https://news.ycombinator.com/item?id=8885392

etc etc.

Do elaborate, please.

Hell banning serves a purpose. If you see someone who has been hell banned you check their post hostory. Sometimes the hellban is accidental in which case you let the person onow. In other cases it's probably deserved, in which case you either let the person know but remind them of the site guidelines or you don't let them know.

This poster has a bunch of posts that shouldn't be on HN.

Then those posts should be blocked, not his good ones. Sorry, but this guy is not abusive. And "hellbanning" is just the creepiest way to moderate a forum.

[+] Digital Ocean

    Pros 1. Inexpensive, Pay as you go
         2. automated deployment
    Cons 1. Capped Network I/O

[+] Azure

    Pros 1. Cheap CDN, all-in-one cloud solutions
         2. Pay as you go
    Cons 1. Expensive VM's (0.6gb $13)

[3] RunAbove

    Pros 1. Inexpensive
    Cons 1. Not mature yet, they changed their pricing twice already

Linode. Support is great, service is great, pricing is good, and seems less "cowboyish" than Digital Ocean and others.

Yeah, using Linode as well. Surprised it isn't more popular TBH.

And the pricing is almost equivalent to Digital Ocean at this point, so long as you're not looking to spin instances up and down for periods less than a month.

I've been running a VPN over multiple data centres on Linode for the past few months. Seems to work great so far. The performance is good enough for what I'm doing with it and I know exactly what my bill is going to be.

Microsoft Azure

- The admin portal is great

- Deploying to websites is just a simple push

- I can programmatically spin up new SQL DB instances

- I really like C#

- There are a LOT of additional services available

- Microsoft is very aggressive about keeping prices low

I think Azure gets a bad reputation in this community from the old "Microsoft = bad" bias that a lot of us have. However, if you are in a Windows stack and/or a corporate developer (and that is a large but quiet percentage of developers) it really seems like the best solution. Also the variety of services they provide under one umbrella is neck and neck with Amazon for best in the industry.

I wanted to try Azure a few weeks ago for a small project. But I found the UI to be confusing. Part old and part new ? Trying to match up with the current Windows look (which I am not familiar with) ?

I also had trouble using one of their API's. Got even thrown out very quickly for abusing it. No idea why, just trying to access. Maybe MS does not like my coding style :-) Never had something like that before. Reference to possible explanation page did not work. Faq did not cover it. Still I got an automated email a few days later to congratulate me with using Azure, while actually I was locked out...

But I have no commitment to Windows, so its easy to switch. I normally use DO and AWS and stick with them for now.

If you really want to manage your Azure you'll have to go PowerShell anyway.

I love the Azure command line tools, but the admin portal last time I tried (Christmas 2013/2014) was 500ing, 404ing, and every other code all the time. It never affected anything since we used the command line tools, but there was a bunch of UI errors that never should have made it into production.

Azure for the main infrastructure - I'm running a .NET MVC app and their Paas offering is excellent.

I agree with the other comments about Azure's VM's being too expensive though.

Amazon S3 and Cloudfront b/c Azure CDN doesn't quite have all the features I need.

AWS - primarily S3 because their cloud storage is inexpensive and easy to administer.

Azure - websites and mobile services (and by extension SQL Database) because we use .NET

AWS, because we make a lot of use of S3 - last I looked there weren't any alternatives that truly rival it. Largely then end up using EC2 for cost (data transfer in-data centre) and billing simplicity.

Used AWS for a couple of years but switch to DO and haven't looked back.

Great interface, pricing and service.

Their customer support is also awesome and I've found their support community articles to be thoroughly useful.

https://www.digitalocean.com/community/

OVH, SoYouStart, and KimSufi with Dokku Alt (for quick apps) or custom Ansible/Docker deployments (for big apps). Since OVH created their North American DC I took the leap and have been very very happy. Knowing just how cheap the hardware/ops game is, I just can't bring myself to spend anything more than what these guys do:

  - $115/mn for a 64GB ECC, 3x Intel SSD, Xeon E51620v2, 500Mbps, etc
  - $50/mn for a 32GB ECC, Intel Xeon W3520, 4TB SATA
  - $15/mn for a 4GB RAM, Atom N2800, 1TB HDD

I scale differently as a result; huge gobs of RAM/SSD/CPU is easily within reach. For instance, I can use Redis Cluster as a primary data store with then op/s an order of magnitude higher than most might be used to.

I used to use Rackspace, but now I'm using AWS. Rackspace was nice, but AWS was a bit better on price (thought the two are priced slightly differently) and a secure setup was quicker on AWS than on Rackspace. I tend to use docs or forums to solve issues I have, so Rackspace's customer support wasn't all that useful compared to whatever Amazon offers.

Since then I've started using a few things like Elastic Beanstalk and continuous deployment, and getting them working on AWS was pretty pain free and easy. I don't think I could do as much on AWS now as I could on Rackspace.

Heroku looks nice but because of price I never tried it. Azure is also tempting to use, but as a non-customer their non-Windows offerings seem to be predictably second-rate.

All of that said, IaaS is still way too complicated. I'm sure there's room for someone to come along and have an offering that's 10x simpler to manage and use.

I have provided my own cloud, in the form of a 1U server, since 1998. I run my (very tiny) website there as well as my own mailserver, and I read email over SSH using (al)pine.

Backups offsite go to the obvious place :)

This setup might not be that interesting, but I think it's worth noting that a cloud computing paradigm/model has functioned well for 16 years now with no significant changes. I have not had data on my own personal computers for that entire 16 years - all data has been in the "cloud".

I think a decent goal is to be a peer on the network. I think a good step in that direction is to have your own server, with fixed IP, on the network. You can build quite a bit on that.

I would love to do this, but handling the security of your own server to me seems labor intensive.

How easy is it for you to manage keeping your server and data secure? I would really like to do this, but my fears of getting hacked have kept me from doing it.

Well, my system runs nothing but sendmail, lighttpd (with no modules or features) and sshd. That's it. So there's not much to lock down.

sshd is hidden behind port knocking, for what that's worth. (No, I do not want to have a religious argument about port knocking this morning)

The real key to the simplicity and security of my setup is email over SSH with a console client. Not only does that remove oceans of attack surface, but it also means you can read email without the emails themselves traversing the network. I guess the characters go over SSH, but that's not quite the same thing...

Thank you for introducing me to port knocking (wikipedia link for the unintiated [0]).

Not trying to lure into an argument you're trying to avoid but I'm just curious; what is the religious argument about regarding port knocking? I know about both sides to tabs/spaces and vim/emacs but am curious what people have against port knocking.

[0] http://en.wikipedia.org/wiki/Port_knocking

Heh.

In short, port knocking is a very, very short/weak password. And is a very weak authentication measure.

This is absolutely true and nobody could argue that.

So if you only did port knocking, or if you depended on port knocking, you're making a bad decision.

I believe in defense in depth, and therefore I think that port knocking on top of everything else you already do has good value - especially considering how simple and lightweight knockd is and my experience of it running stably for years at a time.

Many say it is useless. With passwords + root login disabled to login someone must break your keypair, unlikely to be accomplished by anyone who wouldn't also have the resources to knock ports.

Rackspace

- pricing and offerings are similar to AWS

- they answer the phone when you have problems

Their support is first rate and the #1 reason I won't change to another provider. There is nothing better than opening a chat window and having someone who is knowledgeable respond and who can get the problem fixed right away. To me that is worth the premium I pay for using Rackspace over a service like Azure or AWS.

I've been a big Rackspace Cloud user and advocate here on HN and elsewhere for 6+ years now. This line of thinking was exactly why.

However, in the last year, my experience has been that the quality of support has collapsed. It used to be I could open a chat window and get a knowledgeable person who could fix my problem, but that's not the case anymore. Now I get someone who either doesn't understand what I'm saying or tells me they can't do anything to help unless I upgrade to their gold-plated, top-of-the-line support tier, which would push them from "somewhat more expensive" to "absurdly expensive." It's gotten so bad that I'm in the process of moving all my systems off their cloud at the moment. I can't in good conscience recommend them anymore.

It's all very disappointing, as quality support has been Rackspace's key differentiator forever. My guess is that they've decided to give up competing with AWS on the high end and Digital Ocean et al on the low end, and concentrate on selling into Fortune 500s who need basic VPS services and are willing to pay out the nose for them. But that leaves small shops like mine out in the cold, which is too bad.

Joyent.

+ DTrace available

+ Great predictable performance

+ Solaris

+ no VM

+ fast non-capped network

- client tools are written in node.js (but you can write your own if you really care)

Doesn't matter to me, but might to some people:

- Relatively few datacenters

- Solaris (they do have KVM too, so you can run whatever, but then you lose the many benefits of zones)

Also don't matter to me, but might to some people:

+ completely open source stack, so you could install the same environment in your private datacenter

+ it does not use openstack

I use AWS:

- Everything integrates with it, there's a lot of tooling

- Safety of the herd

Using Azure currently:

- Good PaaS offerings and pretty easy to setup

- Bizspark ($150 free of services per month per account (and yo u can create up to 5 accounts) for 3 years)

- Decent speed (Azure SQL used to be a dog, but has gotten a bit better with the V12 updates)

AWS, because it's what I know to be productive with from previous work. In the early stages of my current business I cannot afford to take the time to learn anything else.

With AWS I know the two biggest risks to its use are vendor lock-in and cost. Vendor lock-in is easy to mitigate early on through wise tech choices, and cost is something to mitigate later at scale when it matters more. And it preserves the option to use those additional, propriety services if at some point it is deemed worth the lock-in risk.

A good first question to ask is what are your needs?

Do you just need to host a web site with some storage? If you're looking at an ecosystem which services interest you?

You might find this article useful: http://www.troyhunt.com/2015/02/stories-from-trenches-sizing...

[Dislaimer: I'm a dev in the Azure Web Apps (was called Azure Websites) team]

Historically, Linode. Good performance, and decent value.

More recently, Heroku. Expensive, but I love being able to leave the sysadmin to someone else.

I set up my first rails app on Linode about 5 years ago and used to scoff at people that went the "easy" route with Heroku.

Now I'm using Heroku and wonder why I wasted so much time before worrying about disk space, permissions, rotating logs, etc. I definitely get it now.

I use RunAbove[1] usually. Mostly because it's nearby (low latency), based on OpenStack (opensource and all the tools works ootb), and inexpensive.

[1] https://www.runabove.com/

Looks too cheap. For how long have you used them and is all your experience positive. Their $/GB RAM is more than twice cheaper than DigitalOcean which are already quite cheap (compared to AWS and the like). Apparent downside of RunAbove is that the ration of cores to memory is lower.

It's OVH, the largest hosting provider in Europe. Cheap is their thing.

I have been using RunAbove specifically since October. It's fine. Even the sandbox instances.

I've used Google Compute Engine to host new web sites. It's cheap, it's rock solid.

I'm still running my fairly large wiki on Linode, just because I haven't had the time to properly migrate it to Compute Engine.

Cycligent.com - layers on top of AWS. Allows for multiple simultaneous versions of an app on the same URL. Also, the easiest deployment I have ever seen.

If I have a standard Rails app then Heroku. For anything more complicated it's either DO or AWS. All services on AWS (R53, S3, SES, etc)

Google Compute Engine. Their Cloud Security team is staffed with rock stars in the industry.

Microsoft Azure

+ Admin portal is good

+ Git Deploy to "Websites" is slick with no setup (builds binaries)

+ NodeJS / C#

+ Azure Integration in Visual Studio

+ Aggressive pricing

The other reason is there are other services that I can use if I don't want to build my own.

Are you posting exactly the same comment multiple times with slightly different wording and a new user account each time?

https://news.ycombinator.com/item?id=9583236

I think the user just copy-pasted the linked comment. The linked comment account is quite old - 2487 days.

No copy and pasting. 2 Different users posting around the same time. We just had same experiences. Mine was originally all in one sentence but I edited it after and reformatted to be bullet points.

The same points, in the same order, in the same hour.

It was within 5 min of me posting. It wasn't displayed when I was typing mine in. I don't have the same points. I didn't even mention SQL DB because I don't use it. I see I also mentioned NodeJS where they didn't. I also talked about Azure integration in Visual Studio and they didn't. These points are quite typical for Azure users.

Reply on news.ycombinator.com