4 ms·
This makes me wonder, how many idle computing resources Amazon has on standby. Are there any numbers on this in the public?
by mgw 16y ago
This makes me wonder, how many idle computing resources Amazon has on standby. Are there any numbers on this in the public?
- BoppreH 16y agoThe thing that always bothered me about cloud computing is where the cloud gets its resources when everybody has a spike, such as during holidays.
- jonknee 16y agoNot everyone gets spikes during holidays--for example Amazon and other retailers get super busy at Christmas, but it's the slowest time of the year for a lot of web apps.
- seldo 16y agoThat said, the last two Decembers we have had trouble instantiating new instances (in US-east-1a, their default and hence most popular location) in December. We solved the problem just by switching to US-east-1b for some tasks.
- influx 16y ago"In order to prevent an overloading of a single availability zone when everybody tries to run their instances in us-east-1a, Amazon has added a layer of indirection so that each account’s availability zones can map to different physical data center equivalents." http://alestic.com/2009/07/ec2-availability-zones http://alestic.com/2009/07/ec2-availability-zones
- ceejayoz 16y ago> The thing that always bothered me about cloud computing is where the cloud gets its resources when everybody has a spike, such as during holidays. When one side of the world is spiking, the other side is sleeping soundly.
- rsynnott 16y agoOccasionally, you just have to wait a while for an instance to start. It's rarely more than half an hour.
- lallysingh 16y agoI recall EC2 going to complete crap during the holidays -- it is (or at least was) their spare christmas capacity.
- tkahn6 16y agoAmazon gave a tech talk at my school. EC2 sprang from the problem that Amazon had to buy a bunch of servers to handle the load around the holidays and these servers went underutilized during the rest of the year. So they decided to lease those resources. When asked about what happens to EC2 during the holidays, the engineer basically replied that Amazon has priority.
- whakojacko 16y agoAccording to Werner Vogels' (Amazon's CTO) Quora answer, the excess capacity story is a myth. http://www.quora.com/How-and-why-did-Amazon-get-into-the-cloud-computing-business http://www.quora.com/How-and-why-did-Amazon-get-into-the-clo...
- tkahn6 16y agoWell then, that's odd. Thanks for clearing that up.
- gojomo 16y agoOr even worse, when some emergent AI suddenly wants all the cycles, all the time. (Will the first sign of a runaway AI be skyrocketing AWS spot prices?)
- riobard 16y agoThis is the beauty of aggregation: when you add a lot more different types of users to the system, the load tends to be balanced out because not everybody needs the same resource at exactly the same time. As long as the distribution of different resource requirement is relatively even, the larger the system is, the stabler it becomes. Statistically, adding multiple standard derivations yield a smaller value than their sum.
- shykes 16y agoNo, that's the theoretical beauty of aggregation. If black friday causes a 10x spike in transactions, a larger system won't help you.
- riobard 16y agoThis kind of edge cases will happen much more frequently when you have multiple smaller systems. A larger system helps you by balancing out those. Will a larger system run out of resources if everyone requests it at the same time? Of course it will, but the likelihood will be statistically smaller than smaller systems.
- wmf 16y agoThanks to the spot market there should be little or no idle capacity.
- nraynaud 16y agomy guess is more like you have waves of computation, with a mass effect of people targeting the cheap spots.