We own our own servers but rent space in a data center. Has increased our CPU/ram exponentially, increased our devops work, increased our risk and time spent on risk mitigation. And decreased our cost over Linode and AWS by an order of magnitude..!
Should I select collocated or in house? The former seems closest but unsure... :). We're happy though!
Dedicated is a nice step in between a
VPS and collocation of your own servers. Plus I had servers near US East that had 10ms or so ping time to MongoHQ which is AWS. So use dedicated for your web/app servers (which are easier to manage) and AWS solutions for your database, search, analytics and monitoring.
I've never managed devops so that is surprising to me. I thought that given the economies of scale, AWS would be cheaper than collocated just because of staffing. How many people do you have employed for devops?
<=1 full-time employee. Our scale isn't huge either, we were running <= 20 decent-sized virtual servers on Linode & AWS previously.
We've definitely been very pleasantly surprised, for sure. The ability to add capacity at very low recurring cost is unbeatable. The communities & OSS that exist now also help to go a pretty solid distance these days -- and we've yet to have any kind of notable issues or catastrophes, though we do have aggressive & tested backup/restore plans in place.
Fingers do remain crossed slightly more-so than they used to when we didn't own the hardware, though. ;-)
VMFarms. I'm a solo-founder, 4 years in business, and just can't do everything anymore. I moved there from Rackspace almost a year ago and absolutely couldn't be happier, they are my devops team now. I have a dev environment on DO for mobile work on the Chromebook too which is working out fine.
My guess would be things are mostly the same aside from DigitalOcean taking some of Linode/Slicehost's least affluent customers. My production stuff for Improvely & W3Counter remain on servers rented from SoftLayer, while my personal sites continue to be hosted on a Linode VPS.
Someone else posted this: http://hnlike.com/hncharts/chart/?id=7791612 which is a useful comparison. Looks like AWS still has the top spot, but what's impressive is how quickly digital ocean has risen to prominence. At the time of writing it's number 2. I use it myself and couldn't be happier with it. Easily one of the easiest to use hosting providers I've used and the technical articles on their site are great as well.
I've used AWS, DO, and most recently Rackspace Cloud for a big client. Rackspace is a great company to deal with if you're willing to pay. Essentially we migrated their servers to the non managed cloud, but at some point in the future if they require they can go to a managed service level and have quick access to great managed talent. I have been pretty happy with the experience so far, the SSD based instances perform faster than the older dedicated hardware they left.
I'm surprised how many co locate their servers. I'm assuming you are reasonably close to the datacenter? In my case the closest data center is a good hour drive. Typically when renting most hosts can swap any part immediately with an identical. I'm curious if people keep an identical parts machine or similar?
Not exactly deployed yet, but it's going to be Linode for sure, perhaps with PiCloud for computation-intensive parts. I looked at AWS for a while, but seems overkill at the moment, besides, the AWS management panel and the pricing pages/simple calculator are not exactly user-friendly – and that might've influenced my decision a tad, having a simpler UI.
For personal and test projects, I use DigitalOcean – the lower price being the main reason.
Linode resizing seems a bit better, too – so for a big project which (hopefully) will need scaling, that's a plus. But I ran a Linode for two months before switching to DO, and tried to resize the disk image, and the whole thing went unresponsive, had to start from scratch (thankfully, no data was lost, I was just playing at that moment).
> perhaps with PiCloud for computation-intensive parts
You know PiCloud is shutting down, and mostly replaced with http://www.multyvac.com/ although they were supposed to launch I think 3 months ago and still haven't. I wish them all the best, but do be wary of basing a product on it.
I'm going to keep saying it but nobody should be using Linode for production use. They have a history of hiding critical issues (in particular security) from their customers. It really is unacceptable and inexcusable conduct.
I've been considering migrating a bunch of things to Linode. I have read a lot of reviews and have not seen too many negative things. Do you have any more information or advice on where I can look to find out about what you are referring to?
I currently use Rackspace for anything critical. My experiences have been extremely positive with them. In a few years of using them I've only had one issue where I wasn't happy with them, but in the hosting world that's pretty good.
The only reason i'm looking to migrate things is because Rackspace is far more expensive.
I guess every provider has its "horror story".
Rackspace suffering outages, DigitalOcean closing a personal site (if I recall correctly) for some dubious reasons, AWS having problems with EBS-EC2 interaction (again, not entirely sure of the details).
I'm still considering Rackspace as an alternative.
Yeah, I know, I did everything as needed. The console showed that it was online, but I couldn't SSH into it or use the LISH web console to access it.
I wrote to their support team, but at that point, I had already redone everything from scratch.
I have 13 production instances on DO and have been very pleased. Some of them are heavy on the CPU. I did quite a bit of testing comparing equivalent DO 8&16GB instances to Linode and DO came out ahead each time. That may have changed now that Linode released all their upgrades.
I'm interested as I have a customer who pays around £200 a month per dedicated box at the minute for nearly half the spec. They are true dedicated boxes though and the network they are on hardly ever goes down (roughly 20 minutes downtime in 3 months due to an edge router firmware upgrade gone badly wrong).
The provider is transparent though and the issue hasn't happened previously or since.
The 16GB RAM package at £100 a month is just amazing if it really is that stable.
Do you do any type of redundancy between DCs / networks? How did you find their redundancy tools if you did?
We had 3 months where we had 20 minutes of downtime. The edge router was ~15 minutes and there was a denial of service on a machine in the same portion of the network which took a few minutes to isolate.
That is the only downtime we received in our 12 month initial contract which started Feb 2013 and ended Feb 2014 (we are on a rolling contract now).
So (20 / 12) = 1.6 minutes a month averaged. For arguments sake it puts us around the four 9s in availability a month.
Same, I have one for a VPN and another one for a simple nodejs app. $5 a month for nodejs hosting with a static IP. The nodejs as a service sites either don't offer external static IPs, or charge a fortune for it. It's been fantastic.
I did it for two reasons really; I wanted to test out DO as a service to see what it had to offer (at $5 a month at my own expense it's written off easy) and I like to learn new things.
I'd done a similar thing on AWS before on their free platform but you only got 100GB a month transfer. This is eaten up quickly on streaming videos.
I'm mainly a .net dev on Windows boxes but I would like to expand my knowledge a little further nginx etc. Most of the open source technologies run best on Linux so I like to dabble in the prompts a little for experience.
I setup an old computer in my home recently and installed Ubuntu as a home media server. The setup ended up costing me £8 in a new RAM stick and about a week of messing to get it working how I wanted. I could have purchased a NAS with it all installed for £200 but I just enjoy the challenge. It's a welcome break from programming sometimes, it's a kind of downtime.
Also, VPN packages start at $9.99 a month if you pay month to month (the savings are the longer term contracts) so the way I see it I am halving that cost and learning at the same time.
I think you're going to see a huge amount of digitaloceans since "production" for me isn't exactly getting a lot of traffic, so it's on a $5 instance.. So I predict digitalocean will rack up a lot of $5 instances and win.
We're running a Rails app on Heroku but we're not happy with its performance/price, although we like the convenience. Is there any of service that rivals Heroku in terms of the convenience of running and maintaining a Rails app?
I recommend Dokku (https://github.com/progrium/dokku) if you want something similar to Heroku. The nice thing about Dokku is that it's simple without locking you in to any particular (proprietary) cloud provider stack.
We converted from Heroku to AWS+Chef (we use hosted chef from opscode). We even wrote and open-sourced a application_procfile cookbook that lets you do heroku-like things with Procfile web and worker processes.
SoftLayer runs good gear, at one point when I worked at a shop where we had almost pure SSD colocated infrastructure, I ran into one of their reps at the bar at a conference and he said that he and I were competing for the same equipment buys from both SuperMicro and in the way of intel SSDs.
We moved from Rackspace to SoftLayer (and running Postgres on bare metal with SSD now), you can hardly get better performance for your money. Generally we're incredibly happy with SL (happy to give you more details if you want, just email me).
No, as far as I know, AWS is the only service that has GPU instances. Probably current demand is not yet high enough to justify the purchase and maintenance of such instances for other cloud providers.
Even PiCloud (which runs on the AWS infrastructure) has no GPU instances.
One problem with Heroku's European region is that it's not Safe Harbor compliant. This means it's not possible to use them and still comply with the data protection laws in some (most?) European countries.
When they announced the European region a year ago they said "Safe Harbor Compliance is Coming Soon" , but there has been no news about it since (as far as I know, at least).
The BizSpark program should be a major draw as well. I just don't think everyone knows about it and the fact that Microsoft will give your startup or even "startup" $150 in Azure credits per month for three years, as well as an ultimate MSDN subscription, both of which can be expanded to multiple accounts for your company or friends so each additional account also gets $150 a month and their own msdn subscription, etc.
I read something the other day that seemed to indicate it's $150 per person in BizSpark. That is, if you're in my company and we both have a BizSpark account we can both make subscriptions on Azure and get the $150 per month _per_ subscription.
We're taking advantage of the $150 and staying within it, but once we go live we'll need to expand. We will definitely investigate that.
Yes, that's what I was trying to get across. Could probably word it better but it's $150 per month, per account you extend your BizSpark account/benefits to.
So if person A signs up for BizSpark, then can then invite x amount of people to their companies BizSpark account and the $150/month and MSDN etc. benefits are also given to those other people. So Person A, B, C, D, etc. will each have their own Bizspark/outlook/azure account and each gets $150 a month in credits, and each has their own ultimate msdn subscription for 3 years.