How and why we switched to Amazon EC2

http://blog.railsonfire.com/2012/10/04/Switching-to-Amazon-EC2.html

by moritzplassnig • 14 years ago

51 49 14 years ago

blog.railsonfire.com

Linode was out of the picture fast, as they do not provide per hour pricing.

I thought that Linode is competitively priced - is that a wrong assumption? (What is the pricing of Amazon's EC2 per year compared to Linode's?)

matlock10• 14 years ago

Linode is very competitively priced, but we need to scale up/down on an hourly basis. when there are less builds to run we simply want less servers to run.

Linode is much cheaper than Amazon when you go over the whole year, but it just doesn't fit our needs.

sp3322• 14 years ago

Linode is cheaper than Amazon, but then every service that charges per-month (not per-hour) is cheaper than Amazon.

ajross2• 14 years ago

CPU time on Linode is cheaper than standard-priced EC2 instances. Compared to a reserved instance they lose badly, though. And last I checked, storage on Linode was 8x higher than EBS.

astrodust6• 14 years ago

Linode charges up-front for the month, but if you terminate an instance early a pro-rated amount is credited to your account. The granularity is at least a day.

aristidb• 14 years ago

Did you consider Reserved Instances?

ajitk3• 14 years ago

Recently stumbled upon very competitively priced DigitalOcean: https://www.digitalocean.com/price-comparison-chart

They have APIs to manage on-demand instances as well.

matlock10• 14 years ago

Digital Ocean looks pretty sweet, but the advantages by going with EC2 for us are a large number of different Instance Types we can use for every task we have.

It provides us with a lot of flexibility on how to build our service.

adanto6840• 14 years ago

Thank you for the link! Their offering looks pretty solid; they are apparently using SSDs for all disk space which is one of the first I've seen doing this while using an 'on-demand' pricing model that's competitive with Linode.

Some key things I saw when I briefly checked it out:

-Their answers [1] regarding bandwidth bursting and fair bandwidth usage / sharing policies [2] is slightly ambiguous, or worrisome at least IMO, although maybe I'm just a cynic heh.

-They do include CPU bursting [3] similar to Linode.

-No private IP addresses [4]. No mention of IPv6 either but perhaps they do support it? =(

-Their 'Features' page looks great but I can't find more specific information & details on some of the things mentioned such as: A) US & Europe data center; is casually mentioned as being New York & Amsterdam only though. B) The ability to upgrade a 'Droplet' VM to a dedicated instance that can use all the host resources. C) No mention of VM resize speed / more info on the 'private network' aspects.

[1] = https://www.digitalocean.com/community/questions/do-you-limi...

[2] = https://www.digitalocean.com/community/questions/do-you-have...

[3] = https://www.digitalocean.com/community/questions/how-does-cp...

[4] = https://www.digitalocean.com/community/questions/do-you-offe...

Does anyone have any first-hand experience with DigitalOcean that could provide some insight?

I'll probably give it a whirl in the next few days as it looks to be a solid offering; would definitely be interested if anyone that may have gone with them already though!

orthecreedence2• 14 years ago

As nice as their offering looks, their TOS prohibits hosting content that is "pornographic, obscene, fraudulent, or discriminatory, including any containing nudity, erotica, profanity, or obscenities" which is extremely restrictive. If you host a website that allows user content, someone posting the word "shit" or posting nude art photography could get your account closed. No thanks.

jimktrains2• 14 years ago

They'll refund unused time.

ezequiel-garzon• 14 years ago

What networking issues did you experience with Hetzner?

matlock10• 14 years ago

Builds failed from time to time because the network was simply gone. We worked around the issues by running commands again when they timed out, but it was one of the reasons (the main reason being that having your own servers is simply not flexible enough) for switching from Hetzner

ashayh• 14 years ago

Network partitioning is also a common occurrence on AWS, and people have to plan for that.

mvkel• 14 years ago

How do you get around the IO issue? We've been looking at EC2 for, well, forever, but IO performance has always been a huge issue -- to the point where it'd cost more than hosting it ourselves just to scale IO performance to the point where it'd equal a couple self-hosted servers.

ajross2• 14 years ago

Not all services are I/O bound. Hell, it's routine in the modern world to have datasets for real world products that fit in, what, $20 of RAM?

matlock10• 14 years ago

Which is exactly the case in our environment. As soon as all libraries are loaded from EBS IO is not that much of an issue any more

matlock10• 14 years ago

We store all the code and everything that changes in a Ram disk, so really fast.

The virtual machine lies on EBS, but it is really fast to start and performance is not limited by EBS at all in our case.

killface• 14 years ago

when i last used ec2, it was almost exclusively for async backend processes. in those cases, high throughput is important, but latency is not a factor in the same way. If there's a 300ms delay talking to the system from the outside world, it's no big deal.

that said, netflix seems to have solved the IO issue with ec2. I'd be curious to know more about their findings.

jedberg• 14 years ago

Main way we solve IO issues: We don't rely on EBS -- most data is stored on the local drives using Cassandra.

For cache throughput, our cache library adds zone affinity to memcache, so the clients are hitting the cache in the same zone first.

Also, we're now exploring their SSD options (http://techblog.netflix.com/2012/07/benchmarking-high-perfor...)

jameskilton• 14 years ago

As it hasn't been mentioned yet, Amazon released EBS Optimized and Provisioned IOPS a few months ago that vastly improves EBS throughput.

http://aws.typepad.com/aws/2012/08/fast-forward-provisioned-...

astrodust6• 14 years ago

Their new SSD-based instances are supposed to be a lot faster, but has anyone benchmarked these?

My experience with EBS was the the performance was decent most of the time, but would brown out intermittently for reasons beyond your control.

peterwwillis3• 14 years ago

If the main motivator here is price (ability to charge per hour vs per month), and EC2 is so much more expensive than a VPS provider, have you done the math to see if buying a couple extra VPS's for the month would not break even or be cheaper overall? Linode is kind of pricey compared to other VPS providers.

The larger your site gets the less you're going to need to rely on "hourly" scaling (as your traffic models also become more predictable). There's also the question of uptime, which we've all seen is not a guarantee from just using Amazon.

grey-area• 14 years ago

From the link:

As a hosted continuous deployment service releasing changes and scaling automatically is a necessity.

I think the key here is that this blog is talking about a hosted service which responds on demand to their client's needs on an hourly basis. EC2 has a lot of infrastructure which does this for them (as long as they can automate it) and is specifically built for this, whereas they'd have to write it themselves by interfacing with a VPS API otherwise. So EC2 might be worth the money to them. VPS providers are mostly more focussed on giving you 1 or 2 boxes to use, and then you leave your site running with occasional adjustments in capacity etc, not on spinning up and down servers on demand, which is a bit more specialised, and might require tens or even hundreds of servers.

For a normal site that's not going to be relevant, and a VPS is cheaper, for this site, it seems EC2 is a better fit...

matlock10• 14 years ago

Exactly. Price itself is not the main issue, but flexibility is.

For example on Weekends we typically see 20%-30% the traffic that we see during the week, simply because people are not at work. Scaling up and down is an absolute necessity there.

peterwwillis3• 14 years ago

I think you're making a lot of assumptions about how a service like their could work and does work. You don't need to write a lot of code to leverage VPS's or scale them, and you don't always need to scale up or down on an hourly basis.

I'll bet you that if they took the historical metrics of their services and their load use and averaged it, buying one or two boxes would easily break even and at the same time not be dependent on a specialized provider or functionality like an auto-scaling VPS.

But that's just a guess.

astrodust6• 14 years ago

> Linode is kind of pricey compared to other VPS providers.

Can you qualify this? The cost is lower than Amazon and Rackspace Cloud.

There are dime-store VPS providers out there that are technically less expensive, but they're usually not able to provide the same level of service. Linode always seems to provide exceptional customer support.

peterwwillis3• 14 years ago

Do you know how many "dime-store VPS providers" there are, and how many of them have existed as long as Linode and longer? Linode provides a valuable, quality service, and has lots of extras that you may not need.

Some VPS providers specialize and have a good history of both uptime and support, and provide it for less than the big-name competitors. I've found many to be exceptionally reliable with good support, and there are communities where you can find reviews and status on such providers to back up their claims.

And in the end, realistically, if you're doing small scaling by the hour you probably don't need the biggest providers in the game. A small service provider for a small company is often a great fit.

You can check out reviews on http://www.lowendbox.com/ and http://www.webhostingtalk.com/ and compare prices, services and reputation. In general, a standard small-sized VPS will be half the cost of Linode, meaning you can buy double the VPS's (on different providers, for example) and get double the capacity and redundancy for the same cost. If you spread them out by region you can implement cheap software GeoIP load-balancing and serve the content by servers closest to your users.

orthecreedence2• 14 years ago

Not to mention Linode is the only VPS provider I've seen (and I've seen quite a few) to offer Slackware boxes. For someone who grew up using Slack, this is a huge plus.

matlock10• 14 years ago

As I was asked about the implementation details I'll write a second blogpost at the end of the week that goes into the details of our implementation.

mp99e99• 14 years ago

Dont forget www.atlantic.net/cloud fast io competitive price per sec billing API etc

How and why we switched to Amazon EC2

Related Stories