Amazon EC2 Reserved Instances now offer instance size flexibility
aws.amazon.com
aws.amazon.com
One more thing - can you make it easier for International customers (I am in Australia) to on sell instances that are surplus to needs on the marketplace? I found out too late that there is a requirement for a US Bank account in order to sell spare or unused reserved instances, which is a PITA and affects our long term planning.
I work in the NT, and we have our own self hosted servers using tomcat as the servlet container. Eg.. Spring/Hibernate/ ect.. Although I find it hard to see how you could leverage those heavy weight systems (gig's in memory to Amazon's EC2?
Some of our web app URLs are in my bio still, I believe (I don't want to repost here in case it is considered promotional spam). They are mainly Ruby/MySQL type apps and the memory requirements are not high, but we need uptime availability and bandwidth, plus ability to scale up and down as demand changes.
Feel free to contact me on Twitter @dsabar if you want to discuss in depth.
Actually you raise an interesting point - The price of bandwidth here in Australia is a moot point now, because only about 5% of our servers are within the AWS Oceania datacentre. All the rest are in the US data centres because most of our applications are applicable to a worldwide audience, and we wanted minimum latency for our biggest target market.
EC2 costs are by far the biggest cost in our monthly billing. Easily running up to 80% of the total monthly costs. The rest is split fairly equally between RDS, CloudFront, DynamoDB, Route53, S3, Glacier etc. Bandwidth itself is a pretty tiny proportion. But then again, we don't do things like transport big files around to/from clients etc. Most of our web apps are purely screen and transactional information.
The biggest cost saving to us in the past few years was moving from on demand instance pricing to reserved instance pricing. Easily knocked our monthly bill down by more than half.
The tricky thing about reserved instances though (at least up to now), was trying to predict what sort of capacity you would need for the next 12 months or so. We did inadvertently spend too much sometimes on 'wrong' sized reserved instances, and basically threw money away because we ended up paying out months on instances that we could not immediately use.
Once we get better at crystal ball gazing, we will probably move to 36 month advance purchase of instances to save even more money.
Oh - one thing that tricked us too - it is not readily apparent on the Amazon Console Reserve Instance purchase screen which instance types work within a VPC and which don't. We ended up reserving a few instance types that we eventually found out we could not use within our defined VPC, which was a shame, though not a deal killer as we could use their "ClassicLink" feature to work around it. Just a tip for any Amazon team member who may be reading.
If you spent so much on ec2, it really worth to take a look at Google cloud compute 3-years term. In high-cpu set the vCPU is cheaper(less than $12 per vcpu), and there is no upfront fee. What stopped me the most from AWS 3-years RI is the huge upfront fee.
If anyone else knows, that'd be helpful - I'm thinking of just buying a ton of *.smalls...
// Edit: Oops, maybe I misunderstood the parent. Sibling's right; you'd need 256 small RIs to make up for one 32xlarge RI. I'm wondering why we should ever buy 32xlarge RIs and lose granularity.
The bigger win, for me, is the AZ flex, as it is a pain to manage multi-AZ setups.
First with Regional RIs I no longer have to worry about AZ and modifying RIs, now I don't worry about RI modification to have right size.
I pay $400/month for a half rack at a datacenter. Its regional RIs also provide instance size flexibility in addition to AZ flexibility so you no longer have to worry about being tied to a specific size or AZ, or worry about launching the right instance size in the right AZ to match their RIs... or something. $400/month.
I mean ffs, I would be laughed out of the room if I went to my investors and said we were putting everything in a datacenter. The historical statistics are that datacenters are far more likely to be a risk to the business than AWS is. It's as simple as that.
Out of curiosity, do you really think that operating cost is the only metric that matters from a business standpoint?
Also, something like EC2 gives you much more support than a half rack in a datacenter does. You need to look beyond the system specs.
Even Google's Cloud (which I don't use and don't promote) has managed to integrate long-term use discounts into their pricing structure so that you don't have to engage in craziness like this.
And FWIW, I would much rather deal with the complexities of a datacenter than specialize in the equal or greater complexities required to manage proprietary, black-hole solutions like this that serve to lock you in by forcing you to specialize in something that is not translatable to other service providers.
By and large, this is going to make life a lot easier for most of my clients.
That said:
* You still have to watch out for generational changes ("Oh, you have a lot of reservations for i2 instances? That sucks, i3s just came out, are a third cheaper, and you can't use your RIs for them."), so 1 year is still where it's at unless you're doing something like writing your cloud spend as CapEx.
* You still need to figure out how your application profiles. Converting an m4 reservation to a c3 isn't happening.
* Smaller instance reservations apply to larger instances, and pro-rate the difference down to on-demand pricing. This is "the right thing," but it's going to make Amazon's already byzantine billing even harder to dissect.